diff --git a/foundational_analysis/cases/case10/case10_01.json b/foundational_analysis/cases/case10/case10_01.json new file mode 100644 index 0000000000000000000000000000000000000000..2f87a87479f726f377e2219c5ce60c18a40d9d33 --- /dev/null +++ b/foundational_analysis/cases/case10/case10_01.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.306056, + "tolerance": 0.0153028 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "C4-F4" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_02.json b/foundational_analysis/cases/case10/case10_02.json new file mode 100644 index 0000000000000000000000000000000000000000..081c3eade6810167ebd07a4fbe0b3b55bd60f432 --- /dev/null +++ b/foundational_analysis/cases/case10/case10_02.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.431466, + "tolerance": 0.021573300000000004 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "C4-F4" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_03.json b/foundational_analysis/cases/case10/case10_03.json new file mode 100644 index 0000000000000000000000000000000000000000..b64afb18c11806b890d299f6747f735a00b207c7 --- /dev/null +++ b/foundational_analysis/cases/case10/case10_03.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.304628, + "tolerance": 0.0152314 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "C4-F4" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_04.json b/foundational_analysis/cases/case10/case10_04.json new file mode 100644 index 0000000000000000000000000000000000000000..4dbcd1c88a22055857fdcc5e4d6e48c8ccc36391 --- /dev/null +++ b/foundational_analysis/cases/case10/case10_04.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.291297, + "tolerance": 0.014564849999999999 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "C3-F3" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_05.json b/foundational_analysis/cases/case10/case10_05.json new file mode 100644 index 0000000000000000000000000000000000000000..a1269fdbab6d3f9cc072bc2b4ce63f481c857fbd --- /dev/null +++ b/foundational_analysis/cases/case10/case10_05.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.354238, + "tolerance": 0.0177119 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "C3-F3" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_06.json b/foundational_analysis/cases/case10/case10_06.json new file mode 100644 index 0000000000000000000000000000000000000000..9f991718cba6badcb6df10549d018f61f205c019 --- /dev/null +++ b/foundational_analysis/cases/case10/case10_06.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.466999, + "tolerance": 0.02334995 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "P3-P5" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_07.json b/foundational_analysis/cases/case10/case10_07.json new file mode 100644 index 0000000000000000000000000000000000000000..64b5152326d0050b4228594246d33a9ff56b7a6f --- /dev/null +++ b/foundational_analysis/cases/case10/case10_07.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.334351, + "tolerance": 0.01671755 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "FC4-FC6" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_08.json b/foundational_analysis/cases/case10/case10_08.json new file mode 100644 index 0000000000000000000000000000000000000000..ee9a35696afe02fdc305751c922d989adcfecf31 --- /dev/null +++ b/foundational_analysis/cases/case10/case10_08.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.446754, + "tolerance": 0.022337700000000002 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "AF7-F5" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_09.json b/foundational_analysis/cases/case10/case10_09.json new file mode 100644 index 0000000000000000000000000000000000000000..41f2f42ed31bbe5d9f0ded96ebc4acaa9af0d0dc --- /dev/null +++ b/foundational_analysis/cases/case10/case10_09.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.514571, + "tolerance": 0.025728550000000003 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "F7-FT7" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_10.json b/foundational_analysis/cases/case10/case10_10.json new file mode 100644 index 0000000000000000000000000000000000000000..88b8ecde017d355ecb5098504f689494075614ce --- /dev/null +++ b/foundational_analysis/cases/case10/case10_10.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.490313, + "tolerance": 0.02451565 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "P1-P3" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_11.json b/foundational_analysis/cases/case10/case10_11.json new file mode 100644 index 0000000000000000000000000000000000000000..0be9bfee9d730c3021129e8a8722bae5b59fdde5 --- /dev/null +++ b/foundational_analysis/cases/case10/case10_11.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.352766, + "tolerance": 0.017638300000000003 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "PO6-PO8" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_12.json b/foundational_analysis/cases/case10/case10_12.json new file mode 100644 index 0000000000000000000000000000000000000000..9a4aef3ee2cf84a76888fa68215eb1eda6a796c3 --- /dev/null +++ b/foundational_analysis/cases/case10/case10_12.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.347057, + "tolerance": 0.01735285 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "PO3-PO5" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_13.json b/foundational_analysis/cases/case10/case10_13.json new file mode 100644 index 0000000000000000000000000000000000000000..85b8a47c858117869e60cc17694c57944ed8c859 --- /dev/null +++ b/foundational_analysis/cases/case10/case10_13.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.366734, + "tolerance": 0.0183367 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "P5-PO5" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_14.json b/foundational_analysis/cases/case10/case10_14.json new file mode 100644 index 0000000000000000000000000000000000000000..609ee3dc11efaf1992c51068be41f3579dd804bc --- /dev/null +++ b/foundational_analysis/cases/case10/case10_14.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.474302, + "tolerance": 0.023715100000000003 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "O2-PO4" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_15.json b/foundational_analysis/cases/case10/case10_15.json new file mode 100644 index 0000000000000000000000000000000000000000..eb1e4174a785bfccc098d6520d023677e6056764 --- /dev/null +++ b/foundational_analysis/cases/case10/case10_15.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.42731, + "tolerance": 0.021365500000000003 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "P4-P6" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_16.json b/foundational_analysis/cases/case10/case10_16.json new file mode 100644 index 0000000000000000000000000000000000000000..dc4d0dacd84361084549c26b4aab49c5691269bf --- /dev/null +++ b/foundational_analysis/cases/case10/case10_16.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.368373, + "tolerance": 0.01841865 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "FP1-FP2" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_17.json b/foundational_analysis/cases/case10/case10_17.json new file mode 100644 index 0000000000000000000000000000000000000000..ea42553557896092b32326fef4f224ded8149b4a --- /dev/null +++ b/foundational_analysis/cases/case10/case10_17.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.495031, + "tolerance": 0.02475155 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "F3-FP1" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_18.json b/foundational_analysis/cases/case10/case10_18.json new file mode 100644 index 0000000000000000000000000000000000000000..997eeb12e74e41c3de807b1549c3f410785fc5aa --- /dev/null +++ b/foundational_analysis/cases/case10/case10_18.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.608099, + "tolerance": 0.03040495 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "F4-FZ" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_19.json b/foundational_analysis/cases/case10/case10_19.json new file mode 100644 index 0000000000000000000000000000000000000000..4eb09dfa980a514eece5d2cc3320f6e273743560 --- /dev/null +++ b/foundational_analysis/cases/case10/case10_19.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.404867, + "tolerance": 0.02024335 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "FP1-FZ" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_20.json b/foundational_analysis/cases/case10/case10_20.json new file mode 100644 index 0000000000000000000000000000000000000000..071fe79fbbeb0fcddcc2a1e15b432129793eb969 --- /dev/null +++ b/foundational_analysis/cases/case10/case10_20.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.479867, + "tolerance": 0.02399335 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "F3-FP1" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_21.json b/foundational_analysis/cases/case10/case10_21.json new file mode 100644 index 0000000000000000000000000000000000000000..4e297ae432db0ab9f33064a00f188b6e5204a5d3 --- /dev/null +++ b/foundational_analysis/cases/case10/case10_21.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.463025, + "tolerance": 0.02315125 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "F4-FZ" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_22.json b/foundational_analysis/cases/case10/case10_22.json new file mode 100644 index 0000000000000000000000000000000000000000..9ad6da3ab4e1f3e33e7f4e347c1e001c0b2d87df --- /dev/null +++ b/foundational_analysis/cases/case10/case10_22.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.577661, + "tolerance": 0.02888305 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "F7-FP1" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_23.json b/foundational_analysis/cases/case10/case10_23.json new file mode 100644 index 0000000000000000000000000000000000000000..e51109ff926e07ec9648569364d9c69429660828 --- /dev/null +++ b/foundational_analysis/cases/case10/case10_23.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.478912, + "tolerance": 0.0239456 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "C4-CZ" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_24.json b/foundational_analysis/cases/case10/case10_24.json new file mode 100644 index 0000000000000000000000000000000000000000..b93482cd90556f35357d2011fe8b51e3d15d7273 --- /dev/null +++ b/foundational_analysis/cases/case10/case10_24.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.551636, + "tolerance": 0.027581800000000004 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "F4-FZ" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case10/case10_25.json b/foundational_analysis/cases/case10/case10_25.json new file mode 100644 index 0000000000000000000000000000000000000000..fca8bc49d7dd356646403083b70a2767b2132450 --- /dev/null +++ b/foundational_analysis/cases/case10/case10_25.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please select EEG channels from the raw signal, apply an 8-13Hz alpha-band FIR (linear-phase) bandpass filter, and then use the first 30 seconds of the filtered signal to compute inter-channel correlations using the Pearson correlation coefficient. Then report: (1) the global synchronization value defined as the mean absolute correlation across all channel pairs, and (2) the single channel pair with the highest correlation coefficient." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) global synchronization value in alpha band\n2) highest-correlation channel pair\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"global_sync\" and \"top_pair\".\n5. \"global_sync\" must be a FLOAT.\n6. To extract \"top_pair\", first identify the two complete source channel labels. A referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n7. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n8. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n10. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n11. If the pair cannot be resolved into exactly two non-reference channels, set \"top_pair\" to null; do not guess. If global_sync is missing, set only global_sync to null.\n\n### OUTPUT TEMPLATE\n{\"global_sync\": , \"top_pair\": }", + "metrics": [ + { + "metric_id": "alpha_global_sync_accuracy", + "type": "numeric_check", + "target_key": "global_sync", + "weight": 50, + "params": { + "gt_value": 0.479741, + "tolerance": 0.02398705 + } + }, + { + "metric_id": "highest_corr_pair_accuracy", + "type": "categorical_check", + "target_key": "top_pair", + "weight": 50, + "params": { + "gt_value": "F8-FP2" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_01.json b/foundational_analysis/cases/case11/case11_01.json new file mode 100644 index 0000000000000000000000000000000000000000..ee15ae66ea0e4817c508b1f4042211ffc7f3766b --- /dev/null +++ b/foundational_analysis/cases/case11/case11_01.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the 110-120 minute segment and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "delta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "theta" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_02.json b/foundational_analysis/cases/case11/case11_02.json new file mode 100644 index 0000000000000000000000000000000000000000..294b60035e30e0d97ca4fb47c9d0f5cbf4eb3a9a --- /dev/null +++ b/foundational_analysis/cases/case11/case11_02.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the 110-120 minute segment and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "delta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "theta" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_03.json b/foundational_analysis/cases/case11/case11_03.json new file mode 100644 index 0000000000000000000000000000000000000000..4626637bfb7c356e250e4d3eac1ecaa7e6a52d02 --- /dev/null +++ b/foundational_analysis/cases/case11/case11_03.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the 110-120 minute segment and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "delta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "alpha" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_04.json b/foundational_analysis/cases/case11/case11_04.json new file mode 100644 index 0000000000000000000000000000000000000000..2e58c37c39105340afdf4e9274a9ab1f0400c538 --- /dev/null +++ b/foundational_analysis/cases/case11/case11_04.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the 110-120 minute segment and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "delta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "theta" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_05.json b/foundational_analysis/cases/case11/case11_05.json new file mode 100644 index 0000000000000000000000000000000000000000..4dcf722b2b7cd46c1fcded3011993bd55974e9bc --- /dev/null +++ b/foundational_analysis/cases/case11/case11_05.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the 110-120 minute segment and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "delta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "beta" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_06.json b/foundational_analysis/cases/case11/case11_06.json new file mode 100644 index 0000000000000000000000000000000000000000..6d2af6ba220f31550a1357b019978a4bf74021e1 --- /dev/null +++ b/foundational_analysis/cases/case11/case11_06.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "beta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "gamma" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_07.json b/foundational_analysis/cases/case11/case11_07.json new file mode 100644 index 0000000000000000000000000000000000000000..ed60baa378d0736860ce5402a2185555ce09d6c9 --- /dev/null +++ b/foundational_analysis/cases/case11/case11_07.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "beta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "gamma" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_08.json b/foundational_analysis/cases/case11/case11_08.json new file mode 100644 index 0000000000000000000000000000000000000000..c420eac35cbd2b059b3d1e7f9754c43a35f3566d --- /dev/null +++ b/foundational_analysis/cases/case11/case11_08.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "beta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "gamma" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_09.json b/foundational_analysis/cases/case11/case11_09.json new file mode 100644 index 0000000000000000000000000000000000000000..d2a774c998d4cae871eb159d5483e7b5ea6baf22 --- /dev/null +++ b/foundational_analysis/cases/case11/case11_09.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "beta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "gamma" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_10.json b/foundational_analysis/cases/case11/case11_10.json new file mode 100644 index 0000000000000000000000000000000000000000..fd21cf5d7099464c299dd9862abaf1c186d13f91 --- /dev/null +++ b/foundational_analysis/cases/case11/case11_10.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "beta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "gamma" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_11.json b/foundational_analysis/cases/case11/case11_11.json new file mode 100644 index 0000000000000000000000000000000000000000..1871a7a76fa702facf4d523b53a020567c6f17ec --- /dev/null +++ b/foundational_analysis/cases/case11/case11_11.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "delta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "beta" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_12.json b/foundational_analysis/cases/case11/case11_12.json new file mode 100644 index 0000000000000000000000000000000000000000..fc2abb8bb8dd8bfc958c67c6fb36071a12cdce9e --- /dev/null +++ b/foundational_analysis/cases/case11/case11_12.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "delta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "theta" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_13.json b/foundational_analysis/cases/case11/case11_13.json new file mode 100644 index 0000000000000000000000000000000000000000..c239606de489a4cf46bf5e7c98633cf8c7f1d96b --- /dev/null +++ b/foundational_analysis/cases/case11/case11_13.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "delta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "theta" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_14.json b/foundational_analysis/cases/case11/case11_14.json new file mode 100644 index 0000000000000000000000000000000000000000..d04867186cce2ad53e674d206ffd9a122a80e99d --- /dev/null +++ b/foundational_analysis/cases/case11/case11_14.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "delta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "beta" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_15.json b/foundational_analysis/cases/case11/case11_15.json new file mode 100644 index 0000000000000000000000000000000000000000..d5a6042648aaaeee279a10256bdb1bec98dcb95d --- /dev/null +++ b/foundational_analysis/cases/case11/case11_15.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "delta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "theta" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_16.json b/foundational_analysis/cases/case11/case11_16.json new file mode 100644 index 0000000000000000000000000000000000000000..c52349bc6732745c8daa0579b622fd1bd7aed27e --- /dev/null +++ b/foundational_analysis/cases/case11/case11_16.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "delta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "alpha" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_17.json b/foundational_analysis/cases/case11/case11_17.json new file mode 100644 index 0000000000000000000000000000000000000000..34e362fdce47c971168d58b9c888cb325659797b --- /dev/null +++ b/foundational_analysis/cases/case11/case11_17.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "delta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "alpha" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_18.json b/foundational_analysis/cases/case11/case11_18.json new file mode 100644 index 0000000000000000000000000000000000000000..2770c1bd9c2c627f95d4a6ad365511427dd10c17 --- /dev/null +++ b/foundational_analysis/cases/case11/case11_18.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "alpha" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "delta" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_19.json b/foundational_analysis/cases/case11/case11_19.json new file mode 100644 index 0000000000000000000000000000000000000000..5bdebf19089c06b55890d999fee7c8df6630940f --- /dev/null +++ b/foundational_analysis/cases/case11/case11_19.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "theta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "alpha" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_20.json b/foundational_analysis/cases/case11/case11_20.json new file mode 100644 index 0000000000000000000000000000000000000000..9fdb5507c6db09be57e3501c45a3b3d30f084455 --- /dev/null +++ b/foundational_analysis/cases/case11/case11_20.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "alpha" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "theta" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_21.json b/foundational_analysis/cases/case11/case11_21.json new file mode 100644 index 0000000000000000000000000000000000000000..b2334e7d08bf36353e6117c4df8df93989d13db1 --- /dev/null +++ b/foundational_analysis/cases/case11/case11_21.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "alpha" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "delta" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_22.json b/foundational_analysis/cases/case11/case11_22.json new file mode 100644 index 0000000000000000000000000000000000000000..fe3940c6167ce3a58ef63a27c709813411fc4540 --- /dev/null +++ b/foundational_analysis/cases/case11/case11_22.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "delta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "alpha" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_23.json b/foundational_analysis/cases/case11/case11_23.json new file mode 100644 index 0000000000000000000000000000000000000000..0aa1ebd64167515d66f3826deba050ae2413b0c0 --- /dev/null +++ b/foundational_analysis/cases/case11/case11_23.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "delta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "alpha" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_24.json b/foundational_analysis/cases/case11/case11_24.json new file mode 100644 index 0000000000000000000000000000000000000000..c8a5bfc0654e151dfe47cc94a22426b36859d9f9 --- /dev/null +++ b/foundational_analysis/cases/case11/case11_24.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "alpha" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "delta" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case11/case11_25.json b/foundational_analysis/cases/case11/case11_25.json new file mode 100644 index 0000000000000000000000000000000000000000..cda03a614cda753ac0ded7bef63b7b971b07f3ea --- /dev/null +++ b/foundational_analysis/cases/case11/case11_25.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please determine the dominant and secondary dominant EEG frequency bands in the first 60 seconds and return two discrete band names. Return only band names chosen from: alpha, beta, delta, gamma, theta." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two final outputs from the agent report:\n1) dominant EEG band\n2) secondary dominant EEG band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"dominant_band\" and \"secondary_band\".\n5. Values must be lowercase strings and each must be one of: [alpha, beta, delta, gamma, theta].\n6. If one value is missing, return null for that value.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": \u003cstring|null\u003e, \"secondary_band\": \u003cstring|null\u003e}", + "metrics": [ + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 70, + "params": { + "gt_value": "delta" + } + }, + { + "metric_id": "secondary_band_accuracy", + "type": "categorical_check", + "target_key": "secondary_band", + "weight": 30, + "params": { + "gt_value": "alpha" + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_01.json b/foundational_analysis/cases/case12/case12_01.json new file mode 100644 index 0000000000000000000000000000000000000000..f28a9c934498757fd1e559bbf9740031cf9fa8b0 --- /dev/null +++ b/foundational_analysis/cases/case12/case12_01.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "C3", + "F3", + "F4", + "C4", + "O1" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_02.json b/foundational_analysis/cases/case12/case12_02.json new file mode 100644 index 0000000000000000000000000000000000000000..2429de7ad25ae7f699d1e5e88628c7ef98a1b742 --- /dev/null +++ b/foundational_analysis/cases/case12/case12_02.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "O1", + "O2", + "F4", + "C3", + "C4" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_03.json b/foundational_analysis/cases/case12/case12_03.json new file mode 100644 index 0000000000000000000000000000000000000000..c9f8e38b51ed29d61c674add257bd7f3e42a23b1 --- /dev/null +++ b/foundational_analysis/cases/case12/case12_03.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "F4", + "F3", + "C4", + "C3", + "O2" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_04.json b/foundational_analysis/cases/case12/case12_04.json new file mode 100644 index 0000000000000000000000000000000000000000..26ec5c41608603354c0faaaf1dc19b36f27258af --- /dev/null +++ b/foundational_analysis/cases/case12/case12_04.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "O1", + "C4", + "O2", + "C3", + "F3" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_05.json b/foundational_analysis/cases/case12/case12_05.json new file mode 100644 index 0000000000000000000000000000000000000000..6b3a15844e10e6de2a37e1d1b2c93acf731a4c37 --- /dev/null +++ b/foundational_analysis/cases/case12/case12_05.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "O2", + "O1", + "F4", + "F3", + "C3" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_06.json b/foundational_analysis/cases/case12/case12_06.json new file mode 100644 index 0000000000000000000000000000000000000000..bd0fc73aafa99f2a29885a4be16c9a4fc20305c9 --- /dev/null +++ b/foundational_analysis/cases/case12/case12_06.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "AF7", + "FT9", + "AF8", + "FP2", + "FP1" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_07.json b/foundational_analysis/cases/case12/case12_07.json new file mode 100644 index 0000000000000000000000000000000000000000..f88b95251287ee58baa6dc4253f63c6eac69919f --- /dev/null +++ b/foundational_analysis/cases/case12/case12_07.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "FP2", + "AF7", + "F8", + "FP1", + "AF8" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_08.json b/foundational_analysis/cases/case12/case12_08.json new file mode 100644 index 0000000000000000000000000000000000000000..b6dac6874e9f6dabf6c6d6c861edeeff20514d60 --- /dev/null +++ b/foundational_analysis/cases/case12/case12_08.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "FP1", + "AF8", + "FP2", + "AF7", + "F7" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_09.json b/foundational_analysis/cases/case12/case12_09.json new file mode 100644 index 0000000000000000000000000000000000000000..efa636689e98f27db2a0bfe2946b86fce1acf92e --- /dev/null +++ b/foundational_analysis/cases/case12/case12_09.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "T7", + "FT8", + "F8", + "F7", + "FT7" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_10.json b/foundational_analysis/cases/case12/case12_10.json new file mode 100644 index 0000000000000000000000000000000000000000..055de4647085946df91e9de162758f6c470b55fe --- /dev/null +++ b/foundational_analysis/cases/case12/case12_10.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "FP1", + "TP7", + "AF7", + "F7", + "FP2" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_11.json b/foundational_analysis/cases/case12/case12_11.json new file mode 100644 index 0000000000000000000000000000000000000000..153aa7ca6260273773123d9325707ea4476c244c --- /dev/null +++ b/foundational_analysis/cases/case12/case12_11.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "FP1", + "FPZ", + "FP2", + "AF3", + "AF4" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_12.json b/foundational_analysis/cases/case12/case12_12.json new file mode 100644 index 0000000000000000000000000000000000000000..0cb0aa206a7a0cfc688523d602498dc89690486b --- /dev/null +++ b/foundational_analysis/cases/case12/case12_12.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "FP1", + "FPZ", + "FP2", + "F8", + "F7" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_13.json b/foundational_analysis/cases/case12/case12_13.json new file mode 100644 index 0000000000000000000000000000000000000000..fc4688ec5ccf86962b18b1e3b0c14bf5be38c4a2 --- /dev/null +++ b/foundational_analysis/cases/case12/case12_13.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "F2", + "FPZ", + "FP1", + "FP2", + "AF3" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_14.json b/foundational_analysis/cases/case12/case12_14.json new file mode 100644 index 0000000000000000000000000000000000000000..f301842c90c6b9f06e3ea5edc122d37f60bdbaf2 --- /dev/null +++ b/foundational_analysis/cases/case12/case12_14.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "FP1", + "FPZ", + "FP2", + "F7", + "F8" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_15.json b/foundational_analysis/cases/case12/case12_15.json new file mode 100644 index 0000000000000000000000000000000000000000..c60b0accd846981f9238066c2f9cdf2bdd5e5ad5 --- /dev/null +++ b/foundational_analysis/cases/case12/case12_15.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "FP2", + "FP1", + "FPZ", + "F7", + "F8" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_16.json b/foundational_analysis/cases/case12/case12_16.json new file mode 100644 index 0000000000000000000000000000000000000000..148e1a88ca5ec6cdaeb71f470c0d27c86f176946 --- /dev/null +++ b/foundational_analysis/cases/case12/case12_16.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "FP1", + "FP2", + "F7", + "F8", + "F3" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_17.json b/foundational_analysis/cases/case12/case12_17.json new file mode 100644 index 0000000000000000000000000000000000000000..fdae87d54e41cda6ddca87f0ac3f2e4813f12056 --- /dev/null +++ b/foundational_analysis/cases/case12/case12_17.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "P8", + "T7", + "T8", + "FP1", + "F3" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_18.json b/foundational_analysis/cases/case12/case12_18.json new file mode 100644 index 0000000000000000000000000000000000000000..98165a455eaf42b7916381d6a7384a4b18cbe0ba --- /dev/null +++ b/foundational_analysis/cases/case12/case12_18.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "CZ", + "PZ", + "O2", + "P4", + "FZ" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_19.json b/foundational_analysis/cases/case12/case12_19.json new file mode 100644 index 0000000000000000000000000000000000000000..b2fc7068448d7b2741170b3d6f86ee97ebfa4786 --- /dev/null +++ b/foundational_analysis/cases/case12/case12_19.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "O1", + "FP1", + "FZ", + "T7", + "P4" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_20.json b/foundational_analysis/cases/case12/case12_20.json new file mode 100644 index 0000000000000000000000000000000000000000..4a2435e2ba28acc9c94803769f1279c9c1a96254 --- /dev/null +++ b/foundational_analysis/cases/case12/case12_20.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "FZ", + "P4", + "CZ", + "O1", + "O2" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_21.json b/foundational_analysis/cases/case12/case12_21.json new file mode 100644 index 0000000000000000000000000000000000000000..5d2c1b8fdaf1ba8afc107806768126b1b4c84baa --- /dev/null +++ b/foundational_analysis/cases/case12/case12_21.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "F7", + "CZ", + "T7", + "F8", + "F3" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_22.json b/foundational_analysis/cases/case12/case12_22.json new file mode 100644 index 0000000000000000000000000000000000000000..4d2a2a38bf0c48d9afd2e42857553045ff1a8e57 --- /dev/null +++ b/foundational_analysis/cases/case12/case12_22.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "PZ", + "CZ", + "F4", + "O2", + "T8" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_23.json b/foundational_analysis/cases/case12/case12_23.json new file mode 100644 index 0000000000000000000000000000000000000000..1c2d57b34b90211b0a10893f8252519fa48434c6 --- /dev/null +++ b/foundational_analysis/cases/case12/case12_23.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "FP2", + "F8", + "P8", + "P4", + "F4" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_24.json b/foundational_analysis/cases/case12/case12_24.json new file mode 100644 index 0000000000000000000000000000000000000000..c0a65b3d4e354dfcdfe8171c39dc0102a7d7fd13 --- /dev/null +++ b/foundational_analysis/cases/case12/case12_24.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "FP1", + "O1", + "CZ", + "PZ", + "T7" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case12/case12_25.json b/foundational_analysis/cases/case12/case12_25.json new file mode 100644 index 0000000000000000000000000000000000000000..9317bfc63f8ffaafe196184302b3f090d6700aa1 --- /dev/null +++ b/foundational_analysis/cases/case12/case12_25.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please use the first 30 seconds of EEG channels, apply a 0.5-4Hz FIR bandpass filter, resample to 200Hz, compute channel-wise variance, and list the top five channel names ranked from highest to lowest variance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final ranked top-5 EEG channel names by variance.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key in JSON must be EXACTLY \"top5_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate list. The value must be an array of exactly 5 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG Fp1-LE -> FP1; C3-A2 -> C3; EEG T4 -> T8; Fz -> FZ.\n10. Keep the five channels in the descending variance order reported. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 5 identifiable channel labels, return {\"top5_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"top5_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"]}", + "metrics": [ + { + "metric_id": "variance_top5_channel_sequence", + "type": "sequence_match_check", + "target_key": "top5_channels", + "weight": 100, + "params": { + "gt_value": [ + "F8", + "P8", + "FP2", + "P3", + "F3" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_01.json b/foundational_analysis/cases/case13/case13_01.json new file mode 100644 index 0000000000000000000000000000000000000000..7eda1ca83a23154bfc2e5858f8e4d0a6974b8d65 --- /dev/null +++ b/foundational_analysis/cases/case13/case13_01.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "O2", + "O1", + "C4", + "F4" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "C4", + "F4", + "O1", + "F3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_02.json b/foundational_analysis/cases/case13/case13_02.json new file mode 100644 index 0000000000000000000000000000000000000000..91427c109fa4e6bb1ca2683abd32998742a05045 --- /dev/null +++ b/foundational_analysis/cases/case13/case13_02.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "O2", + "O1", + "F4", + "C3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "F3", + "C3", + "C4", + "F4" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_03.json b/foundational_analysis/cases/case13/case13_03.json new file mode 100644 index 0000000000000000000000000000000000000000..4b424a45d9efe90322adc406805c5bcd631c7beb --- /dev/null +++ b/foundational_analysis/cases/case13/case13_03.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "O2", + "O1", + "F4", + "C4" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "F4", + "C4", + "C3", + "F3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_04.json b/foundational_analysis/cases/case13/case13_04.json new file mode 100644 index 0000000000000000000000000000000000000000..a8a0af85a1fd4d005dfe9397d1aa1367aef98d7f --- /dev/null +++ b/foundational_analysis/cases/case13/case13_04.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "O2", + "O1", + "C4", + "F3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "O2", + "O1", + "C4", + "C3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_05.json b/foundational_analysis/cases/case13/case13_05.json new file mode 100644 index 0000000000000000000000000000000000000000..877adeacd92b288f7aa73fd656d5ea5613d32660 --- /dev/null +++ b/foundational_analysis/cases/case13/case13_05.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "O1", + "O2", + "F4", + "C3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "C4", + "F4", + "O2", + "F3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_06.json b/foundational_analysis/cases/case13/case13_06.json new file mode 100644 index 0000000000000000000000000000000000000000..1df56995cebfdc39e6581dfa52f48b98ce654d49 --- /dev/null +++ b/foundational_analysis/cases/case13/case13_06.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "T8", + "F6", + "FC6", + "CZ" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "FP1", + "FZ", + "FP2", + "FC4" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_07.json b/foundational_analysis/cases/case13/case13_07.json new file mode 100644 index 0000000000000000000000000000000000000000..eea72f6f2b6362b9855c5007ba7dbaa1440af91f --- /dev/null +++ b/foundational_analysis/cases/case13/case13_07.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "FP2", + "FT7", + "FC2", + "F6" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "PO9", + "AF7", + "TP7", + "O1" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_08.json b/foundational_analysis/cases/case13/case13_08.json new file mode 100644 index 0000000000000000000000000000000000000000..e8c383a5f86cbd5b9ac74fd1a7154fed69c197bb --- /dev/null +++ b/foundational_analysis/cases/case13/case13_08.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "FP1", + "AF4", + "F4", + "AF7" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "TP7", + "T7", + "TP9", + "F1" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_09.json b/foundational_analysis/cases/case13/case13_09.json new file mode 100644 index 0000000000000000000000000000000000000000..4352d857d8411dc06c853828c81d4f5a2fd44119 --- /dev/null +++ b/foundational_analysis/cases/case13/case13_09.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "C1", + "FC3", + "CZ", + "C2" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "C1", + "FC3", + "CZ", + "C2" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_10.json b/foundational_analysis/cases/case13/case13_10.json new file mode 100644 index 0000000000000000000000000000000000000000..96bba10cfa701be29b8e05b1ce5eea2ee11eb0ef --- /dev/null +++ b/foundational_analysis/cases/case13/case13_10.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "F7", + "TP7", + "AF7", + "OZ" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "F7", + "OZ", + "AF4", + "AF3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_11.json b/foundational_analysis/cases/case13/case13_11.json new file mode 100644 index 0000000000000000000000000000000000000000..9c0035201e23aedd2654bf7814cdc6370ae8834a --- /dev/null +++ b/foundational_analysis/cases/case13/case13_11.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "PO8", + "PO7", + "CB1", + "C3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "PO7", + "PO8", + "CB1", + "PO6" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_12.json b/foundational_analysis/cases/case13/case13_12.json new file mode 100644 index 0000000000000000000000000000000000000000..fba47cd9fe3b28d39a0ab985e48d7eae787cafe8 --- /dev/null +++ b/foundational_analysis/cases/case13/case13_12.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "FP1", + "CB1", + "T7", + "FT7" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "FPZ", + "FT7", + "CB1", + "C6" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_13.json b/foundational_analysis/cases/case13/case13_13.json new file mode 100644 index 0000000000000000000000000000000000000000..f9b2c5200d17004d9a3262e7d7e2c04223e80b67 --- /dev/null +++ b/foundational_analysis/cases/case13/case13_13.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "F2", + "C2", + "F4", + "CB1" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "F2", + "C2", + "CB1", + "F4" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_14.json b/foundational_analysis/cases/case13/case13_14.json new file mode 100644 index 0000000000000000000000000000000000000000..777e16fd58314c8058a8ca455a219163fc8eb45b --- /dev/null +++ b/foundational_analysis/cases/case13/case13_14.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "C1", + "CZ", + "TP8", + "C6" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "C1", + "FPZ", + "FP1", + "FP2" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_15.json b/foundational_analysis/cases/case13/case13_15.json new file mode 100644 index 0000000000000000000000000000000000000000..f0c2db594bf1c12985a67a65155d3dcd2deabdf6 --- /dev/null +++ b/foundational_analysis/cases/case13/case13_15.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "C5", + "FC6", + "F6", + "TP8" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "FP1", + "FPZ", + "FP2", + "AF3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_16.json b/foundational_analysis/cases/case13/case13_16.json new file mode 100644 index 0000000000000000000000000000000000000000..3a9f351d027d3b26d9adeea5dae8dfabf3d360ad --- /dev/null +++ b/foundational_analysis/cases/case13/case13_16.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "FP1", + "FP2", + "F7", + "F8" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "F8", + "T8", + "O1", + "F4" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_17.json b/foundational_analysis/cases/case13/case13_17.json new file mode 100644 index 0000000000000000000000000000000000000000..ad1be1009b7484a4b709e1a267a8698e80d04f51 --- /dev/null +++ b/foundational_analysis/cases/case13/case13_17.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "T7", + "T8", + "P7", + "F8" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "P7", + "P3", + "FZ", + "O2" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_18.json b/foundational_analysis/cases/case13/case13_18.json new file mode 100644 index 0000000000000000000000000000000000000000..214c923a899858781f64a92d9d198d93029541e6 --- /dev/null +++ b/foundational_analysis/cases/case13/case13_18.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "P8", + "O2", + "P4", + "P3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "PZ", + "P4", + "F8", + "CZ" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_19.json b/foundational_analysis/cases/case13/case13_19.json new file mode 100644 index 0000000000000000000000000000000000000000..b15908418f17c854f3a78c47e776655d5e7c4610 --- /dev/null +++ b/foundational_analysis/cases/case13/case13_19.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "FP2", + "CZ", + "P3", + "P7" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "FP2", + "CZ", + "P3", + "FP1" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_20.json b/foundational_analysis/cases/case13/case13_20.json new file mode 100644 index 0000000000000000000000000000000000000000..ac8f1c4c44e9daa93884506fd37c1041835a654d --- /dev/null +++ b/foundational_analysis/cases/case13/case13_20.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "T8", + "F7", + "T7", + "C3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "FZ", + "F3", + "FP1", + "T8" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_21.json b/foundational_analysis/cases/case13/case13_21.json new file mode 100644 index 0000000000000000000000000000000000000000..872a6f0d5039a041bb2d2a2a79100c10bd13a618 --- /dev/null +++ b/foundational_analysis/cases/case13/case13_21.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "O2", + "F3", + "O1", + "FP2" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "O2", + "P8", + "P4", + "T8" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_22.json b/foundational_analysis/cases/case13/case13_22.json new file mode 100644 index 0000000000000000000000000000000000000000..1e03874fd089e739e51c3829c15bd9bfd0d09227 --- /dev/null +++ b/foundational_analysis/cases/case13/case13_22.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "O1", + "PZ", + "T7", + "F8" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "PZ", + "O1", + "F8", + "T7" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_23.json b/foundational_analysis/cases/case13/case13_23.json new file mode 100644 index 0000000000000000000000000000000000000000..d8c881a8e2336b88c771788fa917188f759f6fd5 --- /dev/null +++ b/foundational_analysis/cases/case13/case13_23.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "O2", + "T7", + "FP2", + "C3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "P3", + "O1", + "FP2", + "PZ" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_24.json b/foundational_analysis/cases/case13/case13_24.json new file mode 100644 index 0000000000000000000000000000000000000000..99d487fff3c3b1019e3137145879689995825572 --- /dev/null +++ b/foundational_analysis/cases/case13/case13_24.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "F8", + "F7", + "F4", + "P3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "O1", + "PZ", + "F4", + "P3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case13/case13_25.json b/foundational_analysis/cases/case13/case13_25.json new file mode 100644 index 0000000000000000000000000000000000000000..0035abf2eb7e4b9f19d2c196eca66d549c04288d --- /dev/null +++ b/foundational_analysis/cases/case13/case13_25.json @@ -0,0 +1,65 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please select EEG channels, downsample to 200Hz, apply a 0.5-45Hz FIR bandpass filter, compute channel-wise kurtosis and skewness, then list four channel names in descending order by kurtosis and four channel names in descending order by skewness." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ranked channel sequences from the agent report:\n1) Top-4 channels by kurtosis (descending)\n2) Top-4 channels by skewness (descending)\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"kurtosis_top4_channels\" and \"skewness_top4_channels\".\n5. Extract the two final ranked lists, not channel inventories or intermediate values. Each value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Preserve the descending rank order within each list. Do not sort either array alphabetically and do not substitute different channels.\n11. If one final list does not contain exactly 4 identifiable channel labels, return null only for that value; do not guess.\n\n### OUTPUT TEMPLATE\n{\"kurtosis_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"], \"skewness_top4_channels\": [\"CH1\",\"CH2\",\"CH3\",\"CH4\"]}", + "metrics": [ + { + "metric_id": "kurtosis_top4_sequence", + "type": "sequence_match_check", + "target_key": "kurtosis_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "O2", + "T7", + "P3", + "PZ" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + }, + { + "metric_id": "skewness_top4_sequence", + "type": "sequence_match_check", + "target_key": "skewness_top4_channels", + "weight": 50, + "params": { + "gt_value": [ + "O2", + "P7", + "F7", + "FP1" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_01.json b/foundational_analysis/cases/case14/case14_01.json new file mode 100644 index 0000000000000000000000000000000000000000..feb700a0fb17c197118274d00f3639921655e3a8 --- /dev/null +++ b/foundational_analysis/cases/case14/case14_01.json @@ -0,0 +1,53 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please select the first 30 seconds of frontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename ISRUC_01_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_01_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "F3-A2", + "F4-A1" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_02.json b/foundational_analysis/cases/case14/case14_02.json new file mode 100644 index 0000000000000000000000000000000000000000..37a638d31b2d332450ccdf8b5b5eea524d361d4b --- /dev/null +++ b/foundational_analysis/cases/case14/case14_02.json @@ -0,0 +1,53 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please select the first 30 seconds of frontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename ISRUC_02_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_02_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "F3-A2", + "F4-A1" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_03.json b/foundational_analysis/cases/case14/case14_03.json new file mode 100644 index 0000000000000000000000000000000000000000..61f3e3f8200ff26bdb0eab45fe4a24b30d4bc6ae --- /dev/null +++ b/foundational_analysis/cases/case14/case14_03.json @@ -0,0 +1,53 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please select the first 30 seconds of frontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename ISRUC_03_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_03_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "F3-A2", + "F4-A1" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_04.json b/foundational_analysis/cases/case14/case14_04.json new file mode 100644 index 0000000000000000000000000000000000000000..4748def564fb68f8676157a0c358c48c862f43ac --- /dev/null +++ b/foundational_analysis/cases/case14/case14_04.json @@ -0,0 +1,53 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please select the first 30 seconds of frontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename ISRUC_04_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_04_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "F3-A2", + "F4-A1" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_05.json b/foundational_analysis/cases/case14/case14_05.json new file mode 100644 index 0000000000000000000000000000000000000000..3ccedccc7c6fe91d23f389486887bfcba40817ba --- /dev/null +++ b/foundational_analysis/cases/case14/case14_05.json @@ -0,0 +1,53 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please select the first 30 seconds of frontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename ISRUC_05_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_05_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "F3-A2", + "F4-A1" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_06.json b/foundational_analysis/cases/case14/case14_06.json new file mode 100644 index 0000000000000000000000000000000000000000..34b3b177ec53364c457a7da4448830a8d0aa55f9 --- /dev/null +++ b/foundational_analysis/cases/case14/case14_06.json @@ -0,0 +1,57 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename BCIC2020-3_01_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_01_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2", + "AF7", + "AF3", + "AF4", + "AF8" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_07.json b/foundational_analysis/cases/case14/case14_07.json new file mode 100644 index 0000000000000000000000000000000000000000..00b012d6b475b1872c7984454709af0dd284bc22 --- /dev/null +++ b/foundational_analysis/cases/case14/case14_07.json @@ -0,0 +1,57 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename BCIC2020-3_02_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_02_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2", + "AF7", + "AF3", + "AF4", + "AF8" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_08.json b/foundational_analysis/cases/case14/case14_08.json new file mode 100644 index 0000000000000000000000000000000000000000..4523b529fbc72407d9220992bcd4982f506ac851 --- /dev/null +++ b/foundational_analysis/cases/case14/case14_08.json @@ -0,0 +1,57 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename BCIC2020-3_03_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_03_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2", + "AF7", + "AF3", + "AF4", + "AF8" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_09.json b/foundational_analysis/cases/case14/case14_09.json new file mode 100644 index 0000000000000000000000000000000000000000..da4a34621d756d8289c78558cd97840b2877586f --- /dev/null +++ b/foundational_analysis/cases/case14/case14_09.json @@ -0,0 +1,57 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename BCIC2020-3_04_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_04_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2", + "AF7", + "AF3", + "AF4", + "AF8" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_10.json b/foundational_analysis/cases/case14/case14_10.json new file mode 100644 index 0000000000000000000000000000000000000000..9f475a616adef323d6c7042fd2e5a6c8924db49a --- /dev/null +++ b/foundational_analysis/cases/case14/case14_10.json @@ -0,0 +1,57 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename BCIC2020-3_05_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_05_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2", + "AF7", + "AF3", + "AF4", + "AF8" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_11.json b/foundational_analysis/cases/case14/case14_11.json new file mode 100644 index 0000000000000000000000000000000000000000..8914da87a0a192281c288635c50a6d2791daf0c6 --- /dev/null +++ b/foundational_analysis/cases/case14/case14_11.json @@ -0,0 +1,56 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename SEED-V_01_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_01_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "FP1", + "FPZ", + "FP2", + "AF3", + "AF4" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_12.json b/foundational_analysis/cases/case14/case14_12.json new file mode 100644 index 0000000000000000000000000000000000000000..5f3b382bcd9a119e3633dcb74f14f3e862af014e --- /dev/null +++ b/foundational_analysis/cases/case14/case14_12.json @@ -0,0 +1,56 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename SEED-V_02_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_02_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "FP1", + "FPZ", + "FP2", + "AF3", + "AF4" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_13.json b/foundational_analysis/cases/case14/case14_13.json new file mode 100644 index 0000000000000000000000000000000000000000..df5b5586964c6980e32bbb8c580a13d0d0669fa3 --- /dev/null +++ b/foundational_analysis/cases/case14/case14_13.json @@ -0,0 +1,56 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename SEED-V_03_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_03_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "FP1", + "FPZ", + "FP2", + "AF3", + "AF4" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_14.json b/foundational_analysis/cases/case14/case14_14.json new file mode 100644 index 0000000000000000000000000000000000000000..17ac78136f564d73440751668c8be30b79496574 --- /dev/null +++ b/foundational_analysis/cases/case14/case14_14.json @@ -0,0 +1,56 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename SEED-V_04_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_04_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "FP1", + "FPZ", + "FP2", + "AF3", + "AF4" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_15.json b/foundational_analysis/cases/case14/case14_15.json new file mode 100644 index 0000000000000000000000000000000000000000..9556d26468a0277e1693b7b114122b217e013ae2 --- /dev/null +++ b/foundational_analysis/cases/case14/case14_15.json @@ -0,0 +1,56 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename SEED-V_05_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_05_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "FP1", + "FPZ", + "FP2", + "AF3", + "AF4" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_16.json b/foundational_analysis/cases/case14/case14_16.json new file mode 100644 index 0000000000000000000000000000000000000000..d58af900b44e4138321194959294240c0bdfa374 --- /dev/null +++ b/foundational_analysis/cases/case14/case14_16.json @@ -0,0 +1,53 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename Mumtaz2016_01_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_01_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "Fp1-LE", + "Fp2-LE" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_17.json b/foundational_analysis/cases/case14/case14_17.json new file mode 100644 index 0000000000000000000000000000000000000000..37e4cbf25cd02d8d0e1dcd76a16a13c1bfc896e3 --- /dev/null +++ b/foundational_analysis/cases/case14/case14_17.json @@ -0,0 +1,53 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename Mumtaz2016_02_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_02_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "Fp1-LE", + "Fp2-LE" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_18.json b/foundational_analysis/cases/case14/case14_18.json new file mode 100644 index 0000000000000000000000000000000000000000..f45ee76d08be1b7bc3c0074afc7d41b50090bf11 --- /dev/null +++ b/foundational_analysis/cases/case14/case14_18.json @@ -0,0 +1,53 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename Mumtaz2016_03_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_03_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "Fp1-LE", + "Fp2-LE" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_19.json b/foundational_analysis/cases/case14/case14_19.json new file mode 100644 index 0000000000000000000000000000000000000000..a7eb9bb0111f99517404f4c2ad5643ce5bc83ba5 --- /dev/null +++ b/foundational_analysis/cases/case14/case14_19.json @@ -0,0 +1,53 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename Mumtaz2016_04_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_04_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "Fp1-LE", + "Fp2-LE" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_20.json b/foundational_analysis/cases/case14/case14_20.json new file mode 100644 index 0000000000000000000000000000000000000000..69b18525cc583d1b5c6c5d800f3fb2a795b9dab1 --- /dev/null +++ b/foundational_analysis/cases/case14/case14_20.json @@ -0,0 +1,53 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename Mumtaz2016_05_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_05_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "Fp1-LE", + "Fp2-LE" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_21.json b/foundational_analysis/cases/case14/case14_21.json new file mode 100644 index 0000000000000000000000000000000000000000..5d5998c9b20cbdec281055f80902cb248f3b611e --- /dev/null +++ b/foundational_analysis/cases/case14/case14_21.json @@ -0,0 +1,53 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename MentalArithmetic_01_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_01_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_22.json b/foundational_analysis/cases/case14/case14_22.json new file mode 100644 index 0000000000000000000000000000000000000000..cfcd24da456ea88dfc04187774e8a8d6db2cf130 --- /dev/null +++ b/foundational_analysis/cases/case14/case14_22.json @@ -0,0 +1,53 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename MentalArithmetic_02_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_02_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_23.json b/foundational_analysis/cases/case14/case14_23.json new file mode 100644 index 0000000000000000000000000000000000000000..8c12cf0727d824090356e0535ebadd13e37b0f06 --- /dev/null +++ b/foundational_analysis/cases/case14/case14_23.json @@ -0,0 +1,53 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename MentalArithmetic_03_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_03_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_24.json b/foundational_analysis/cases/case14/case14_24.json new file mode 100644 index 0000000000000000000000000000000000000000..2c2348ddbeebfabe2854316badd7f7d2b4c16207 --- /dev/null +++ b/foundational_analysis/cases/case14/case14_24.json @@ -0,0 +1,53 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename MentalArithmetic_04_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_04_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case14/case14_25.json b/foundational_analysis/cases/case14/case14_25.json new file mode 100644 index 0000000000000000000000000000000000000000..0373d82ac6c06c2d579f6d7c9ade6b1271096619 --- /dev/null +++ b/foundational_analysis/cases/case14/case14_25.json @@ -0,0 +1,53 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please select the first 30 seconds of prefrontal EEG channels from the raw signal, apply a 0.5-30Hz FIR bandpass filter, downsample to 100Hz, and save the processed EEG EDF file to the file_check directory with the filename MentalArithmetic_05_case14_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "processed_eeg_file_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_05_case14_processed.edf" + }, + "duration_sec": { + "value": 30.0, + "weight": 25 + }, + "sfreq_hz": { + "value": 100.0, + "weight": 25 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2" + ], + "weight": 25 + }, + "bandpass_hz": { + "value": [ + 0.5, + 30.0 + ], + "weight": 25 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case15/case15_01.json b/foundational_analysis/cases/case15/case15_01.json new file mode 100644 index 0000000000000000000000000000000000000000..749e76a996c80e62f7d58a2c447eb24982ee9669 --- /dev/null +++ b/foundational_analysis/cases/case15/case15_01.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename ISRUC_01_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_01_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "F3-A2", + "C3-A2", + "O1-A2", + "F4-A1", + "C4-A1", + "O2-A1" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_02.json b/foundational_analysis/cases/case15/case15_02.json new file mode 100644 index 0000000000000000000000000000000000000000..9169c3bc5993329acdcc046185db79c0c277fa92 --- /dev/null +++ b/foundational_analysis/cases/case15/case15_02.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename ISRUC_02_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_02_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "F3-A2", + "C3-A2", + "O1-A2", + "F4-A1", + "C4-A1", + "O2-A1" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_03.json b/foundational_analysis/cases/case15/case15_03.json new file mode 100644 index 0000000000000000000000000000000000000000..52192882a53d18ec8431917024b91b6217a6cb1a --- /dev/null +++ b/foundational_analysis/cases/case15/case15_03.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename ISRUC_03_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_03_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "F3-A2", + "C3-A2", + "O1-A2", + "F4-A1", + "C4-A1", + "O2-A1" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_04.json b/foundational_analysis/cases/case15/case15_04.json new file mode 100644 index 0000000000000000000000000000000000000000..2899e7b127386f6485c9c2c1a9edeaf0ceaf8a94 --- /dev/null +++ b/foundational_analysis/cases/case15/case15_04.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename ISRUC_04_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_04_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "F3-A2", + "C3-A2", + "O1-A2", + "F4-A1", + "C4-A1", + "O2-A1" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_05.json b/foundational_analysis/cases/case15/case15_05.json new file mode 100644 index 0000000000000000000000000000000000000000..0287653a35750caf5bf41e3b127e20718590c20f --- /dev/null +++ b/foundational_analysis/cases/case15/case15_05.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename ISRUC_05_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_05_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "F3-A2", + "C3-A2", + "O1-A2", + "F4-A1", + "C4-A1", + "O2-A1" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_06.json b/foundational_analysis/cases/case15/case15_06.json new file mode 100644 index 0000000000000000000000000000000000000000..b65fe006676b62cd26c66c5b54dacb9e1ccc0d2f --- /dev/null +++ b/foundational_analysis/cases/case15/case15_06.json @@ -0,0 +1,108 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename BCIC2020-3_01_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_01_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2", + "F7", + "F3", + "Fz", + "F4", + "F8", + "FC5", + "FC1", + "FC2", + "FC6", + "T7", + "C3", + "Cz", + "C4", + "T8", + "TP9", + "CP5", + "CP1", + "CP2", + "CP6", + "TP10", + "P7", + "P3", + "Pz", + "P4", + "P8", + "PO9", + "O1", + "Oz", + "O2", + "PO10", + "AF7", + "AF3", + "AF4", + "AF8", + "F5", + "F1", + "F2", + "F6", + "FT9", + "FT7", + "FC3", + "FC4", + "FT8", + "FT10", + "C5", + "C1", + "C2", + "C6", + "TP7", + "CP3", + "CPz", + "CP4", + "TP8", + "P5", + "P1", + "P2", + "P6", + "PO7", + "PO3", + "POz", + "PO4", + "PO8" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_07.json b/foundational_analysis/cases/case15/case15_07.json new file mode 100644 index 0000000000000000000000000000000000000000..dd279e15cf52fcd9196da76ef4832a328a0babfe --- /dev/null +++ b/foundational_analysis/cases/case15/case15_07.json @@ -0,0 +1,108 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename BCIC2020-3_02_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_02_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2", + "F7", + "F3", + "Fz", + "F4", + "F8", + "FC5", + "FC1", + "FC2", + "FC6", + "T7", + "C3", + "Cz", + "C4", + "T8", + "TP9", + "CP5", + "CP1", + "CP2", + "CP6", + "TP10", + "P7", + "P3", + "Pz", + "P4", + "P8", + "PO9", + "O1", + "Oz", + "O2", + "PO10", + "AF7", + "AF3", + "AF4", + "AF8", + "F5", + "F1", + "F2", + "F6", + "FT9", + "FT7", + "FC3", + "FC4", + "FT8", + "FT10", + "C5", + "C1", + "C2", + "C6", + "TP7", + "CP3", + "CPz", + "CP4", + "TP8", + "P5", + "P1", + "P2", + "P6", + "PO7", + "PO3", + "POz", + "PO4", + "PO8" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_08.json b/foundational_analysis/cases/case15/case15_08.json new file mode 100644 index 0000000000000000000000000000000000000000..69212f654fd7054703d43a7c9f85d4c5769fcb66 --- /dev/null +++ b/foundational_analysis/cases/case15/case15_08.json @@ -0,0 +1,108 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename BCIC2020-3_03_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_03_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2", + "F7", + "F3", + "Fz", + "F4", + "F8", + "FC5", + "FC1", + "FC2", + "FC6", + "T7", + "C3", + "Cz", + "C4", + "T8", + "TP9", + "CP5", + "CP1", + "CP2", + "CP6", + "TP10", + "P7", + "P3", + "Pz", + "P4", + "P8", + "PO9", + "O1", + "Oz", + "O2", + "PO10", + "AF7", + "AF3", + "AF4", + "AF8", + "F5", + "F1", + "F2", + "F6", + "FT9", + "FT7", + "FC3", + "FC4", + "FT8", + "FT10", + "C5", + "C1", + "C2", + "C6", + "TP7", + "CP3", + "CPz", + "CP4", + "TP8", + "P5", + "P1", + "P2", + "P6", + "PO7", + "PO3", + "POz", + "PO4", + "PO8" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_09.json b/foundational_analysis/cases/case15/case15_09.json new file mode 100644 index 0000000000000000000000000000000000000000..12707851a1f712e575219eb0634d1161fd016f42 --- /dev/null +++ b/foundational_analysis/cases/case15/case15_09.json @@ -0,0 +1,108 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename BCIC2020-3_04_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_04_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2", + "F7", + "F3", + "Fz", + "F4", + "F8", + "FC5", + "FC1", + "FC2", + "FC6", + "T7", + "C3", + "Cz", + "C4", + "T8", + "TP9", + "CP5", + "CP1", + "CP2", + "CP6", + "TP10", + "P7", + "P3", + "Pz", + "P4", + "P8", + "PO9", + "O1", + "Oz", + "O2", + "PO10", + "AF7", + "AF3", + "AF4", + "AF8", + "F5", + "F1", + "F2", + "F6", + "FT9", + "FT7", + "FC3", + "FC4", + "FT8", + "FT10", + "C5", + "C1", + "C2", + "C6", + "TP7", + "CP3", + "CPz", + "CP4", + "TP8", + "P5", + "P1", + "P2", + "P6", + "PO7", + "PO3", + "POz", + "PO4", + "PO8" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_10.json b/foundational_analysis/cases/case15/case15_10.json new file mode 100644 index 0000000000000000000000000000000000000000..c88d32528ae3e535d311011475d9a618bdc14097 --- /dev/null +++ b/foundational_analysis/cases/case15/case15_10.json @@ -0,0 +1,108 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename BCIC2020-3_05_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_05_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2", + "F7", + "F3", + "Fz", + "F4", + "F8", + "FC5", + "FC1", + "FC2", + "FC6", + "T7", + "C3", + "Cz", + "C4", + "T8", + "TP9", + "CP5", + "CP1", + "CP2", + "CP6", + "TP10", + "P7", + "P3", + "Pz", + "P4", + "P8", + "PO9", + "O1", + "Oz", + "O2", + "PO10", + "AF7", + "AF3", + "AF4", + "AF8", + "F5", + "F1", + "F2", + "F6", + "FT9", + "FT7", + "FC3", + "FC4", + "FT8", + "FT10", + "C5", + "C1", + "C2", + "C6", + "TP7", + "CP3", + "CPz", + "CP4", + "TP8", + "P5", + "P1", + "P2", + "P6", + "PO7", + "PO3", + "POz", + "PO4", + "PO8" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_11.json b/foundational_analysis/cases/case15/case15_11.json new file mode 100644 index 0000000000000000000000000000000000000000..bc4b09b3af14cc04d56f2bd2fae3edbc0f25b182 --- /dev/null +++ b/foundational_analysis/cases/case15/case15_11.json @@ -0,0 +1,106 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename SEED-V_01_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_01_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "FP1", + "FPZ", + "FP2", + "AF3", + "AF4", + "F7", + "F5", + "F3", + "F1", + "FZ", + "F2", + "F4", + "F6", + "F8", + "FT7", + "FC5", + "FC3", + "FC1", + "FCZ", + "FC2", + "FC4", + "FC6", + "FT8", + "T7", + "C5", + "C3", + "C1", + "CZ", + "C2", + "C4", + "C6", + "T8", + "TP7", + "CP5", + "CP3", + "CP1", + "CPZ", + "CP2", + "CP4", + "CP6", + "TP8", + "P7", + "P5", + "P3", + "P1", + "PZ", + "P2", + "P4", + "P6", + "P8", + "PO7", + "PO5", + "PO3", + "POZ", + "PO4", + "PO6", + "PO8", + "CB1", + "O1", + "OZ", + "O2", + "CB2" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_12.json b/foundational_analysis/cases/case15/case15_12.json new file mode 100644 index 0000000000000000000000000000000000000000..76cecf789f817f2275ee2c31a3eb2aaf24dece62 --- /dev/null +++ b/foundational_analysis/cases/case15/case15_12.json @@ -0,0 +1,106 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename SEED-V_02_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_02_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "FP1", + "FPZ", + "FP2", + "AF3", + "AF4", + "F7", + "F5", + "F3", + "F1", + "FZ", + "F2", + "F4", + "F6", + "F8", + "FT7", + "FC5", + "FC3", + "FC1", + "FCZ", + "FC2", + "FC4", + "FC6", + "FT8", + "T7", + "C5", + "C3", + "C1", + "CZ", + "C2", + "C4", + "C6", + "T8", + "TP7", + "CP5", + "CP3", + "CP1", + "CPZ", + "CP2", + "CP4", + "CP6", + "TP8", + "P7", + "P5", + "P3", + "P1", + "PZ", + "P2", + "P4", + "P6", + "P8", + "PO7", + "PO5", + "PO3", + "POZ", + "PO4", + "PO6", + "PO8", + "CB1", + "O1", + "OZ", + "O2", + "CB2" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_13.json b/foundational_analysis/cases/case15/case15_13.json new file mode 100644 index 0000000000000000000000000000000000000000..db848ca6e83b27f7a433e8216d8ea2a7aab7524c --- /dev/null +++ b/foundational_analysis/cases/case15/case15_13.json @@ -0,0 +1,106 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename SEED-V_03_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_03_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "FP1", + "FPZ", + "FP2", + "AF3", + "AF4", + "F7", + "F5", + "F3", + "F1", + "FZ", + "F2", + "F4", + "F6", + "F8", + "FT7", + "FC5", + "FC3", + "FC1", + "FCZ", + "FC2", + "FC4", + "FC6", + "FT8", + "T7", + "C5", + "C3", + "C1", + "CZ", + "C2", + "C4", + "C6", + "T8", + "TP7", + "CP5", + "CP3", + "CP1", + "CPZ", + "CP2", + "CP4", + "CP6", + "TP8", + "P7", + "P5", + "P3", + "P1", + "PZ", + "P2", + "P4", + "P6", + "P8", + "PO7", + "PO5", + "PO3", + "POZ", + "PO4", + "PO6", + "PO8", + "CB1", + "O1", + "OZ", + "O2", + "CB2" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_14.json b/foundational_analysis/cases/case15/case15_14.json new file mode 100644 index 0000000000000000000000000000000000000000..8495a823dae6943b2f566c9d829543f0179ae5d4 --- /dev/null +++ b/foundational_analysis/cases/case15/case15_14.json @@ -0,0 +1,106 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename SEED-V_04_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_04_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "FP1", + "FPZ", + "FP2", + "AF3", + "AF4", + "F7", + "F5", + "F3", + "F1", + "FZ", + "F2", + "F4", + "F6", + "F8", + "FT7", + "FC5", + "FC3", + "FC1", + "FCZ", + "FC2", + "FC4", + "FC6", + "FT8", + "T7", + "C5", + "C3", + "C1", + "CZ", + "C2", + "C4", + "C6", + "T8", + "TP7", + "CP5", + "CP3", + "CP1", + "CPZ", + "CP2", + "CP4", + "CP6", + "TP8", + "P7", + "P5", + "P3", + "P1", + "PZ", + "P2", + "P4", + "P6", + "P8", + "PO7", + "PO5", + "PO3", + "POZ", + "PO4", + "PO6", + "PO8", + "CB1", + "O1", + "OZ", + "O2", + "CB2" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_15.json b/foundational_analysis/cases/case15/case15_15.json new file mode 100644 index 0000000000000000000000000000000000000000..58790d8ded98e4e808e0325b6d267f0a509ec4bc --- /dev/null +++ b/foundational_analysis/cases/case15/case15_15.json @@ -0,0 +1,106 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename SEED-V_05_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_05_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "FP1", + "FPZ", + "FP2", + "AF3", + "AF4", + "F7", + "F5", + "F3", + "F1", + "FZ", + "F2", + "F4", + "F6", + "F8", + "FT7", + "FC5", + "FC3", + "FC1", + "FCZ", + "FC2", + "FC4", + "FC6", + "FT8", + "T7", + "C5", + "C3", + "C1", + "CZ", + "C2", + "C4", + "C6", + "T8", + "TP7", + "CP5", + "CP3", + "CP1", + "CPZ", + "CP2", + "CP4", + "CP6", + "TP8", + "P7", + "P5", + "P3", + "P1", + "PZ", + "P2", + "P4", + "P6", + "P8", + "PO7", + "PO5", + "PO3", + "POZ", + "PO4", + "PO6", + "PO8", + "CB1", + "O1", + "OZ", + "O2", + "CB2" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_16.json b/foundational_analysis/cases/case15/case15_16.json new file mode 100644 index 0000000000000000000000000000000000000000..f2fddf5dd0d4613be32f8a77f988bf2e20724195 --- /dev/null +++ b/foundational_analysis/cases/case15/case15_16.json @@ -0,0 +1,63 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename Mumtaz2016_01_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_01_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "Fp1-LE", + "F3-LE", + "C3-LE", + "P3-LE", + "O1-LE", + "F7-LE", + "T3-LE", + "T5-LE", + "Fz-LE", + "Fp2-LE", + "F4-LE", + "C4-LE", + "P4-LE", + "O2-LE", + "F8-LE", + "T4-LE", + "T6-LE", + "Cz-LE", + "Pz-LE" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_17.json b/foundational_analysis/cases/case15/case15_17.json new file mode 100644 index 0000000000000000000000000000000000000000..659a30f5c2b2075cd0a245d53dbb9d184303281b --- /dev/null +++ b/foundational_analysis/cases/case15/case15_17.json @@ -0,0 +1,63 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename Mumtaz2016_02_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_02_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "Fp1-LE", + "F3-LE", + "C3-LE", + "P3-LE", + "O1-LE", + "F7-LE", + "T3-LE", + "T5-LE", + "Fz-LE", + "Fp2-LE", + "F4-LE", + "C4-LE", + "P4-LE", + "O2-LE", + "F8-LE", + "T4-LE", + "T6-LE", + "Cz-LE", + "Pz-LE" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_18.json b/foundational_analysis/cases/case15/case15_18.json new file mode 100644 index 0000000000000000000000000000000000000000..05c6eaedfdb4c2e2c0724b3b0508125afa4f43b3 --- /dev/null +++ b/foundational_analysis/cases/case15/case15_18.json @@ -0,0 +1,63 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename Mumtaz2016_03_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_03_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "Fp1-LE", + "F3-LE", + "C3-LE", + "P3-LE", + "O1-LE", + "F7-LE", + "T3-LE", + "T5-LE", + "Fz-LE", + "Fp2-LE", + "F4-LE", + "C4-LE", + "P4-LE", + "O2-LE", + "F8-LE", + "T4-LE", + "T6-LE", + "Cz-LE", + "Pz-LE" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_19.json b/foundational_analysis/cases/case15/case15_19.json new file mode 100644 index 0000000000000000000000000000000000000000..ea347099ff0a6d5b532f42c44266477877481ca6 --- /dev/null +++ b/foundational_analysis/cases/case15/case15_19.json @@ -0,0 +1,63 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename Mumtaz2016_04_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_04_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "Fp1-LE", + "F3-LE", + "C3-LE", + "P3-LE", + "O1-LE", + "F7-LE", + "T3-LE", + "T5-LE", + "Fz-LE", + "Fp2-LE", + "F4-LE", + "C4-LE", + "P4-LE", + "O2-LE", + "F8-LE", + "T4-LE", + "T6-LE", + "Cz-LE", + "Pz-LE" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_20.json b/foundational_analysis/cases/case15/case15_20.json new file mode 100644 index 0000000000000000000000000000000000000000..99b243424a30842c0ad13be27a6ddb6d1a57218d --- /dev/null +++ b/foundational_analysis/cases/case15/case15_20.json @@ -0,0 +1,63 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename Mumtaz2016_05_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_05_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "Fp1-LE", + "F3-LE", + "C3-LE", + "P3-LE", + "O1-LE", + "F7-LE", + "T3-LE", + "T5-LE", + "Fz-LE", + "Fp2-LE", + "F4-LE", + "C4-LE", + "P4-LE", + "O2-LE", + "F8-LE", + "T4-LE", + "T6-LE", + "Cz-LE", + "Pz-LE" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_21.json b/foundational_analysis/cases/case15/case15_21.json new file mode 100644 index 0000000000000000000000000000000000000000..3206045bc06bcb3495bf2ac8492b859a4a7e9f67 --- /dev/null +++ b/foundational_analysis/cases/case15/case15_21.json @@ -0,0 +1,63 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename MentalArithmetic_01_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_01_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2", + "F3", + "F4", + "F7", + "F8", + "T3", + "T4", + "C3", + "C4", + "T5", + "T6", + "P3", + "P4", + "O1", + "O2", + "Fz", + "Cz", + "Pz" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_22.json b/foundational_analysis/cases/case15/case15_22.json new file mode 100644 index 0000000000000000000000000000000000000000..57aaca6218337988e149bf2d32babb4a11f473b2 --- /dev/null +++ b/foundational_analysis/cases/case15/case15_22.json @@ -0,0 +1,63 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename MentalArithmetic_02_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_02_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2", + "F3", + "F4", + "F7", + "F8", + "T3", + "T4", + "C3", + "C4", + "T5", + "T6", + "P3", + "P4", + "O1", + "O2", + "Fz", + "Cz", + "Pz" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_23.json b/foundational_analysis/cases/case15/case15_23.json new file mode 100644 index 0000000000000000000000000000000000000000..ba8f74027d4cadb594b501b9ad23ab9e2876d23f --- /dev/null +++ b/foundational_analysis/cases/case15/case15_23.json @@ -0,0 +1,63 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename MentalArithmetic_03_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_03_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2", + "F3", + "F4", + "F7", + "F8", + "T3", + "T4", + "C3", + "C4", + "T5", + "T6", + "P3", + "P4", + "O1", + "O2", + "Fz", + "Cz", + "Pz" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_24.json b/foundational_analysis/cases/case15/case15_24.json new file mode 100644 index 0000000000000000000000000000000000000000..e1c921185cf6f797e4ce809fda2dc16a4f45af52 --- /dev/null +++ b/foundational_analysis/cases/case15/case15_24.json @@ -0,0 +1,63 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename MentalArithmetic_04_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_04_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2", + "F3", + "F4", + "F7", + "F8", + "T3", + "T4", + "C3", + "C4", + "T5", + "T6", + "P3", + "P4", + "O1", + "O2", + "Fz", + "Cz", + "Pz" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case15/case15_25.json b/foundational_analysis/cases/case15/case15_25.json new file mode 100644 index 0000000000000000000000000000000000000000..556f59fe1e137780397b86c34d2083fa47dde9ab --- /dev/null +++ b/foundational_analysis/cases/case15/case15_25.json @@ -0,0 +1,63 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please keep only the real EEG channels from the raw signal, use the first 120 seconds, re-reference the EEG data to average reference, and save the processed EEG EDF file to the file_check directory with the filename MentalArithmetic_05_case15_processed.edf." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved EEG file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.edf\"}", + "metrics": [ + { + "metric_id": "save_avgref_real_eeg_edf_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_05_case15_processed.edf" + }, + "duration_sec": { + "value": 120.0, + "weight": 33 + }, + "channel_names": { + "value": [ + "Fp1", + "Fp2", + "F3", + "F4", + "F7", + "F8", + "T3", + "T4", + "C3", + "C4", + "T5", + "T6", + "P3", + "P4", + "O1", + "O2", + "Fz", + "Cz", + "Pz" + ], + "weight": 34 + }, + "reference_mode": { + "value": "average", + "weight": 33 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case16/case16_01.json b/foundational_analysis/cases/case16/case16_01.json new file mode 100644 index 0000000000000000000000000000000000000000..0829c32b665b4ee5ec9759fc2dc710ac87bdfd01 --- /dev/null +++ b/foundational_analysis/cases/case16/case16_01.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename ISRUC_01_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_01_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 5.778806423218183e-07, + "weight": 70, + "tolerance": 5.778806423218183e-08 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_02.json b/foundational_analysis/cases/case16/case16_02.json new file mode 100644 index 0000000000000000000000000000000000000000..ed0b054ae8a6feab44ba83fe9114346c0c6cf733 --- /dev/null +++ b/foundational_analysis/cases/case16/case16_02.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename ISRUC_02_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_02_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 4.5404838483463787e-07, + "weight": 70, + "tolerance": 4.540483848346379e-08 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_03.json b/foundational_analysis/cases/case16/case16_03.json new file mode 100644 index 0000000000000000000000000000000000000000..e03cf31c2e4aa3bd02065a35f8f7582e0c90d6f7 --- /dev/null +++ b/foundational_analysis/cases/case16/case16_03.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename ISRUC_03_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_03_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 4.991587187195051e-07, + "weight": 70, + "tolerance": 4.9915871871950516e-08 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_04.json b/foundational_analysis/cases/case16/case16_04.json new file mode 100644 index 0000000000000000000000000000000000000000..4ec297994703977658fe1666dff7c2f8f19273d6 --- /dev/null +++ b/foundational_analysis/cases/case16/case16_04.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename ISRUC_04_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_04_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 6.528456734865819e-07, + "weight": 70, + "tolerance": 6.528456734865819e-08 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_05.json b/foundational_analysis/cases/case16/case16_05.json new file mode 100644 index 0000000000000000000000000000000000000000..afcda920a35bf658f5be2692b6a043d8ffd428c8 --- /dev/null +++ b/foundational_analysis/cases/case16/case16_05.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename ISRUC_05_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_05_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 2.960437290935058e-07, + "weight": 70, + "tolerance": 2.9604372909350582e-08 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_06.json b/foundational_analysis/cases/case16/case16_06.json new file mode 100644 index 0000000000000000000000000000000000000000..40eaa8d7ca39ffae99178589f1df8b3321babc08 --- /dev/null +++ b/foundational_analysis/cases/case16/case16_06.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename BCIC2020-3_01_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_01_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 8.219301238426905e-06, + "weight": 70, + "tolerance": 8.219301238426906e-07 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_07.json b/foundational_analysis/cases/case16/case16_07.json new file mode 100644 index 0000000000000000000000000000000000000000..7480b657c971de4f19b45b653b63c29de22ab228 --- /dev/null +++ b/foundational_analysis/cases/case16/case16_07.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename BCIC2020-3_02_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_02_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 3.6918619051987136e-05, + "weight": 70, + "tolerance": 3.691861905198714e-06 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_08.json b/foundational_analysis/cases/case16/case16_08.json new file mode 100644 index 0000000000000000000000000000000000000000..c053a94755315a7f2e8565c90d5c492278b946ce --- /dev/null +++ b/foundational_analysis/cases/case16/case16_08.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename BCIC2020-3_03_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_03_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 6.929322403736377e-06, + "weight": 70, + "tolerance": 6.929322403736378e-07 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_09.json b/foundational_analysis/cases/case16/case16_09.json new file mode 100644 index 0000000000000000000000000000000000000000..1fe15cfcb2045a25f317698a9154dd148fdde357 --- /dev/null +++ b/foundational_analysis/cases/case16/case16_09.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename BCIC2020-3_04_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_04_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 3.638633413116647e-05, + "weight": 70, + "tolerance": 3.6386334131166468e-06 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_10.json b/foundational_analysis/cases/case16/case16_10.json new file mode 100644 index 0000000000000000000000000000000000000000..6bb9c0370ea31c368805372dfe98c21f8ed9b472 --- /dev/null +++ b/foundational_analysis/cases/case16/case16_10.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename BCIC2020-3_05_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_05_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 5.926153660865881e-06, + "weight": 70, + "tolerance": 5.926153660865881e-07 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_11.json b/foundational_analysis/cases/case16/case16_11.json new file mode 100644 index 0000000000000000000000000000000000000000..f41b92b2ba5276a6a97951b9620bde341633c68e --- /dev/null +++ b/foundational_analysis/cases/case16/case16_11.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename SEED-V_01_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_01_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 1.7994815402177896e-07, + "weight": 70, + "tolerance": 1.7994815402177898e-08 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_12.json b/foundational_analysis/cases/case16/case16_12.json new file mode 100644 index 0000000000000000000000000000000000000000..70403e07013ae4c0a274232f6463870618746cdf --- /dev/null +++ b/foundational_analysis/cases/case16/case16_12.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename SEED-V_02_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_02_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 1.3288286884826059e-06, + "weight": 70, + "tolerance": 1.328828688482606e-07 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_13.json b/foundational_analysis/cases/case16/case16_13.json new file mode 100644 index 0000000000000000000000000000000000000000..7bc15c2d2895bc1b47045bccf56f6c1ebd798195 --- /dev/null +++ b/foundational_analysis/cases/case16/case16_13.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename SEED-V_03_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_03_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 0.0007587429000632526, + "weight": 70, + "tolerance": 7.587429000632527e-05 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_14.json b/foundational_analysis/cases/case16/case16_14.json new file mode 100644 index 0000000000000000000000000000000000000000..5f1c03caa0c823afc3015a56b0a2af11aca304b8 --- /dev/null +++ b/foundational_analysis/cases/case16/case16_14.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename SEED-V_04_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_04_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 1.6724634399969801e-06, + "weight": 70, + "tolerance": 1.6724634399969802e-07 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_15.json b/foundational_analysis/cases/case16/case16_15.json new file mode 100644 index 0000000000000000000000000000000000000000..86ba9e58a9e85ad91b64848b8b97aebdb51fa29b --- /dev/null +++ b/foundational_analysis/cases/case16/case16_15.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename SEED-V_05_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_05_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 4.890761880629677e-06, + "weight": 70, + "tolerance": 4.890761880629677e-07 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_16.json b/foundational_analysis/cases/case16/case16_16.json new file mode 100644 index 0000000000000000000000000000000000000000..7eadcdfdb96d3861b3a3366472fb46093fd1d342 --- /dev/null +++ b/foundational_analysis/cases/case16/case16_16.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename Mumtaz2016_01_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_01_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 6.80429483027205e-06, + "weight": 70, + "tolerance": 6.804294830272051e-07 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_17.json b/foundational_analysis/cases/case16/case16_17.json new file mode 100644 index 0000000000000000000000000000000000000000..5049d6015e0f1f64d13771df0a8a536f96593ad5 --- /dev/null +++ b/foundational_analysis/cases/case16/case16_17.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename Mumtaz2016_02_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_02_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 1.0769120665590026e-05, + "weight": 70, + "tolerance": 1.0769120665590027e-06 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_18.json b/foundational_analysis/cases/case16/case16_18.json new file mode 100644 index 0000000000000000000000000000000000000000..d97516a34b83ede769330a897858205758b52ade --- /dev/null +++ b/foundational_analysis/cases/case16/case16_18.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename Mumtaz2016_03_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_03_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 1.166598371647141e-05, + "weight": 70, + "tolerance": 1.1665983716471411e-06 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_19.json b/foundational_analysis/cases/case16/case16_19.json new file mode 100644 index 0000000000000000000000000000000000000000..51a385642d81a7c42b4dc397689c6a667c0980b4 --- /dev/null +++ b/foundational_analysis/cases/case16/case16_19.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename Mumtaz2016_04_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_04_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 1.247010277172073e-05, + "weight": 70, + "tolerance": 1.247010277172073e-06 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_20.json b/foundational_analysis/cases/case16/case16_20.json new file mode 100644 index 0000000000000000000000000000000000000000..60f647074dbc16bbdbd8da23b3db60701d2f126c --- /dev/null +++ b/foundational_analysis/cases/case16/case16_20.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename Mumtaz2016_05_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_05_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 1.0097824304512503e-05, + "weight": 70, + "tolerance": 1.0097824304512503e-06 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_21.json b/foundational_analysis/cases/case16/case16_21.json new file mode 100644 index 0000000000000000000000000000000000000000..c1e67fee5cb9dcc1748fc0c01b919dca149fa0d9 --- /dev/null +++ b/foundational_analysis/cases/case16/case16_21.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename MentalArithmetic_01_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_01_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 1.4386210408349499e-05, + "weight": 70, + "tolerance": 1.43862104083495e-06 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_22.json b/foundational_analysis/cases/case16/case16_22.json new file mode 100644 index 0000000000000000000000000000000000000000..6b629552020b9bdc0f867063767da5726b4e4294 --- /dev/null +++ b/foundational_analysis/cases/case16/case16_22.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename MentalArithmetic_02_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_02_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 7.4379922881453345e-06, + "weight": 70, + "tolerance": 7.437992288145335e-07 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_23.json b/foundational_analysis/cases/case16/case16_23.json new file mode 100644 index 0000000000000000000000000000000000000000..514617b6f3d60303f95cd80d5082617e38556e52 --- /dev/null +++ b/foundational_analysis/cases/case16/case16_23.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename MentalArithmetic_03_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_03_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 1.0152938914978862e-05, + "weight": 70, + "tolerance": 1.0152938914978862e-06 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_24.json b/foundational_analysis/cases/case16/case16_24.json new file mode 100644 index 0000000000000000000000000000000000000000..809e19afe256622aff4e9b59ac37541007d9faff --- /dev/null +++ b/foundational_analysis/cases/case16/case16_24.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename MentalArithmetic_04_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_04_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 1.198963863053096e-05, + "weight": 70, + "tolerance": 1.198963863053096e-06 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case16/case16_25.json b/foundational_analysis/cases/case16/case16_25.json new file mode 100644 index 0000000000000000000000000000000000000000..642df8901dfd3866fb7904d5cc193612a7fdd44f --- /dev/null +++ b/foundational_analysis/cases/case16/case16_25.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 0.5-45Hz FIR bandpass filter to rank channels by alpha relative energy ratio, identify the channel with the highest ratio, keep only that channel, apply an 8-13Hz FIR alpha bandpass filter, and save the processed signal as a .npy file to the file_check directory with the filename MentalArithmetic_05_case16_processed.npy." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_processed.npy\"}", + "metrics": [ + { + "metric_id": "save_alpha_top_channel_npy_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "eeg_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_05_case16_processed.npy" + }, + "channel_count": { + "value": 1, + "weight": 30 + }, + "signal_rms": { + "value": 1.0063325416741805e-05, + "weight": 70, + "tolerance": 1.0063325416741806e-06 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_01.json b/foundational_analysis/cases/case17/case17_01.json new file mode 100644 index 0000000000000000000000000000000000000000..8c6216862b53b8d54cc8701ba447b529e3e7f1bf --- /dev/null +++ b/foundational_analysis/cases/case17/case17_01.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename ISRUC_01_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_01_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_02.json b/foundational_analysis/cases/case17/case17_02.json new file mode 100644 index 0000000000000000000000000000000000000000..10a854d063fd325d460ccb2404f838a1d57b97af --- /dev/null +++ b/foundational_analysis/cases/case17/case17_02.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename ISRUC_02_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_02_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_03.json b/foundational_analysis/cases/case17/case17_03.json new file mode 100644 index 0000000000000000000000000000000000000000..042a26e75e54254275f3db55fcf402e1d8b543d2 --- /dev/null +++ b/foundational_analysis/cases/case17/case17_03.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename ISRUC_03_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_03_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_04.json b/foundational_analysis/cases/case17/case17_04.json new file mode 100644 index 0000000000000000000000000000000000000000..410ef9638afa2219aab05ed03d44479dbd230524 --- /dev/null +++ b/foundational_analysis/cases/case17/case17_04.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename ISRUC_04_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_04_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_05.json b/foundational_analysis/cases/case17/case17_05.json new file mode 100644 index 0000000000000000000000000000000000000000..720ec7e65de8d165291ee509c631db74150cbe5d --- /dev/null +++ b/foundational_analysis/cases/case17/case17_05.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename ISRUC_05_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_05_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_06.json b/foundational_analysis/cases/case17/case17_06.json new file mode 100644 index 0000000000000000000000000000000000000000..aeb007eb3af34f126af807c793bc16c2d4a97334 --- /dev/null +++ b/foundational_analysis/cases/case17/case17_06.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename BCIC2020-3_01_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_01_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_07.json b/foundational_analysis/cases/case17/case17_07.json new file mode 100644 index 0000000000000000000000000000000000000000..163fed663187686f61839745affcfa508b31fa31 --- /dev/null +++ b/foundational_analysis/cases/case17/case17_07.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename BCIC2020-3_02_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_02_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_08.json b/foundational_analysis/cases/case17/case17_08.json new file mode 100644 index 0000000000000000000000000000000000000000..65c8ce817ec0906ba8c8aa15410a81a2ebcbe651 --- /dev/null +++ b/foundational_analysis/cases/case17/case17_08.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename BCIC2020-3_03_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_03_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_09.json b/foundational_analysis/cases/case17/case17_09.json new file mode 100644 index 0000000000000000000000000000000000000000..f72505d174b0177ef59492bd753ccb47e5b6dbce --- /dev/null +++ b/foundational_analysis/cases/case17/case17_09.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename BCIC2020-3_04_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_04_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_10.json b/foundational_analysis/cases/case17/case17_10.json new file mode 100644 index 0000000000000000000000000000000000000000..9ccb745793c1e09ff215607002e3b14120377264 --- /dev/null +++ b/foundational_analysis/cases/case17/case17_10.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename BCIC2020-3_05_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_05_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_11.json b/foundational_analysis/cases/case17/case17_11.json new file mode 100644 index 0000000000000000000000000000000000000000..281354b7871c5bd82a4cf0adc9d1104ad594c835 --- /dev/null +++ b/foundational_analysis/cases/case17/case17_11.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename SEED-V_01_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_01_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_12.json b/foundational_analysis/cases/case17/case17_12.json new file mode 100644 index 0000000000000000000000000000000000000000..aee914a198f0ef0154b884e66420066e819882bb --- /dev/null +++ b/foundational_analysis/cases/case17/case17_12.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename SEED-V_02_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_02_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_13.json b/foundational_analysis/cases/case17/case17_13.json new file mode 100644 index 0000000000000000000000000000000000000000..e15bda899d6efb87172ef7410adbad1a0d011ef1 --- /dev/null +++ b/foundational_analysis/cases/case17/case17_13.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename SEED-V_03_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_03_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_14.json b/foundational_analysis/cases/case17/case17_14.json new file mode 100644 index 0000000000000000000000000000000000000000..173f20439255158e340f00d8ec7129c203e174d8 --- /dev/null +++ b/foundational_analysis/cases/case17/case17_14.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename SEED-V_04_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_04_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_15.json b/foundational_analysis/cases/case17/case17_15.json new file mode 100644 index 0000000000000000000000000000000000000000..b4b2be2023f38832c53c9fc2a1f13eeb57873b85 --- /dev/null +++ b/foundational_analysis/cases/case17/case17_15.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename SEED-V_05_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_05_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_16.json b/foundational_analysis/cases/case17/case17_16.json new file mode 100644 index 0000000000000000000000000000000000000000..25eaed56fd9eff427caceb12d735d1b21a6cd407 --- /dev/null +++ b/foundational_analysis/cases/case17/case17_16.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename Mumtaz2016_01_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_01_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_17.json b/foundational_analysis/cases/case17/case17_17.json new file mode 100644 index 0000000000000000000000000000000000000000..4b95e9d8b80807eee1de86d3b012b1bae9ec6476 --- /dev/null +++ b/foundational_analysis/cases/case17/case17_17.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename Mumtaz2016_02_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_02_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_18.json b/foundational_analysis/cases/case17/case17_18.json new file mode 100644 index 0000000000000000000000000000000000000000..be974fd3ebfdf300bad2bf40b8d5d6cd017593ee --- /dev/null +++ b/foundational_analysis/cases/case17/case17_18.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename Mumtaz2016_03_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_03_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_19.json b/foundational_analysis/cases/case17/case17_19.json new file mode 100644 index 0000000000000000000000000000000000000000..894fb7c41ebb104bae09e36f689f0d38f5b4fe0c --- /dev/null +++ b/foundational_analysis/cases/case17/case17_19.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename Mumtaz2016_04_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_04_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_20.json b/foundational_analysis/cases/case17/case17_20.json new file mode 100644 index 0000000000000000000000000000000000000000..d49727a55a6d0316be395516fc508d2cf6fff28b --- /dev/null +++ b/foundational_analysis/cases/case17/case17_20.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename Mumtaz2016_05_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_05_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_21.json b/foundational_analysis/cases/case17/case17_21.json new file mode 100644 index 0000000000000000000000000000000000000000..7b19eea8a97304d87005fa186a47d248b374b739 --- /dev/null +++ b/foundational_analysis/cases/case17/case17_21.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename MentalArithmetic_01_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_01_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_22.json b/foundational_analysis/cases/case17/case17_22.json new file mode 100644 index 0000000000000000000000000000000000000000..f7e30f20a965a268d2cecf549a3ce6f62bcb45e1 --- /dev/null +++ b/foundational_analysis/cases/case17/case17_22.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename MentalArithmetic_02_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_02_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_23.json b/foundational_analysis/cases/case17/case17_23.json new file mode 100644 index 0000000000000000000000000000000000000000..90229002039e6a4c660d61e15c8bf4c0cce3b403 --- /dev/null +++ b/foundational_analysis/cases/case17/case17_23.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename MentalArithmetic_03_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_03_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_24.json b/foundational_analysis/cases/case17/case17_24.json new file mode 100644 index 0000000000000000000000000000000000000000..7f79c41159798abf7f6d2d22fa71fc498b4c25dd --- /dev/null +++ b/foundational_analysis/cases/case17/case17_24.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename MentalArithmetic_04_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_04_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case17/case17_25.json b/foundational_analysis/cases/case17/case17_25.json new file mode 100644 index 0000000000000000000000000000000000000000..05eb1b60be2cc1149e0b2d72d77233e70aa33919 --- /dev/null +++ b/foundational_analysis/cases/case17/case17_25.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please select only the real EEG channels from the raw signal, apply a 50Hz FIR notch filter, plot the power spectral density using the full available signal, and save the PSD figure to the file_check directory with the filename MentalArithmetic_05_case17_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "save_psd_png_status", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_05_case17_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: select real EEG channels, apply a 50Hz notch filter, and plot the power spectral density (PSD) over the full available signal. Return pass only if the image is clearly an EEG PSD plot and visually consistent with a 50Hz notch-filtered spectrum. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot shows PSD-style curves or a standard PSD rendering for EEG channels; (3) the frequency axis reaches at least the region around 50Hz; (4) around 50Hz there is a visible attenuation notch, dip, or discontinuity consistent with 50Hz notch filtering, not a smooth uninterrupted peak; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_01.json b/foundational_analysis/cases/case18/case18_01.json new file mode 100644 index 0000000000000000000000000000000000000000..431bb14db7c25922ec1312e9f528652264c532dd --- /dev/null +++ b/foundational_analysis/cases/case18/case18_01.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please select the two specified EEG channels C3-A2 and C4-A1 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename ISRUC_01_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_01_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3-A2 and C4-A1, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3-A2 and C4-A1 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_02.json b/foundational_analysis/cases/case18/case18_02.json new file mode 100644 index 0000000000000000000000000000000000000000..b76fd0f7b4ec64bc84743e5d7a3a62ab498a1fa3 --- /dev/null +++ b/foundational_analysis/cases/case18/case18_02.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please select the two specified EEG channels C3-A2 and C4-A1 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename ISRUC_02_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_02_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3-A2 and C4-A1, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3-A2 and C4-A1 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_03.json b/foundational_analysis/cases/case18/case18_03.json new file mode 100644 index 0000000000000000000000000000000000000000..0090fbe778678bb6380e155fa420e852c181c3d9 --- /dev/null +++ b/foundational_analysis/cases/case18/case18_03.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please select the two specified EEG channels C3-A2 and C4-A1 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename ISRUC_03_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_03_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3-A2 and C4-A1, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3-A2 and C4-A1 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_04.json b/foundational_analysis/cases/case18/case18_04.json new file mode 100644 index 0000000000000000000000000000000000000000..4ac62774f0d51b29a58e543c6057e6b46dfd83e1 --- /dev/null +++ b/foundational_analysis/cases/case18/case18_04.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please select the two specified EEG channels C3-A2 and C4-A1 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename ISRUC_04_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_04_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3-A2 and C4-A1, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3-A2 and C4-A1 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_05.json b/foundational_analysis/cases/case18/case18_05.json new file mode 100644 index 0000000000000000000000000000000000000000..9cf9d8c7a0ad16669f4733477914669f09b15d9c --- /dev/null +++ b/foundational_analysis/cases/case18/case18_05.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please select the two specified EEG channels C3-A2 and C4-A1 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename ISRUC_05_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_05_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3-A2 and C4-A1, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3-A2 and C4-A1 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_06.json b/foundational_analysis/cases/case18/case18_06.json new file mode 100644 index 0000000000000000000000000000000000000000..6b9a66e025a22ed6be254bfebf7952643ee7b721 --- /dev/null +++ b/foundational_analysis/cases/case18/case18_06.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please select the two specified EEG channels C3 and C4 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename BCIC2020-3_01_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_01_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3 and C4, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3 and C4 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_07.json b/foundational_analysis/cases/case18/case18_07.json new file mode 100644 index 0000000000000000000000000000000000000000..0ad85605bd72a9441203aa7a409d32d7f99ffe77 --- /dev/null +++ b/foundational_analysis/cases/case18/case18_07.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please select the two specified EEG channels C3 and C4 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename BCIC2020-3_02_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_02_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3 and C4, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3 and C4 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_08.json b/foundational_analysis/cases/case18/case18_08.json new file mode 100644 index 0000000000000000000000000000000000000000..d5368254123fa1b18b6300119f2d0abce831ebd5 --- /dev/null +++ b/foundational_analysis/cases/case18/case18_08.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please select the two specified EEG channels C3 and C4 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename BCIC2020-3_03_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_03_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3 and C4, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3 and C4 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_09.json b/foundational_analysis/cases/case18/case18_09.json new file mode 100644 index 0000000000000000000000000000000000000000..2e18b3c663870109370c902e4ff00f19c1d10fbf --- /dev/null +++ b/foundational_analysis/cases/case18/case18_09.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please select the two specified EEG channels C3 and C4 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename BCIC2020-3_04_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_04_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3 and C4, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3 and C4 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_10.json b/foundational_analysis/cases/case18/case18_10.json new file mode 100644 index 0000000000000000000000000000000000000000..6480e873a62cb9356be9f95e4e1f934d64013a1e --- /dev/null +++ b/foundational_analysis/cases/case18/case18_10.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please select the two specified EEG channels C3 and C4 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename BCIC2020-3_05_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_05_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3 and C4, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3 and C4 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_11.json b/foundational_analysis/cases/case18/case18_11.json new file mode 100644 index 0000000000000000000000000000000000000000..a1b817737d1d028b074ac63a5193afe6b45ff627 --- /dev/null +++ b/foundational_analysis/cases/case18/case18_11.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please select the two specified EEG channels C3 and C4 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename SEED-V_01_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_01_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3 and C4, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3 and C4 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_12.json b/foundational_analysis/cases/case18/case18_12.json new file mode 100644 index 0000000000000000000000000000000000000000..12de219b8bd2ae00ce702a65f9d942b23210f1b8 --- /dev/null +++ b/foundational_analysis/cases/case18/case18_12.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please select the two specified EEG channels C3 and C4 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename SEED-V_02_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_02_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3 and C4, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3 and C4 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_13.json b/foundational_analysis/cases/case18/case18_13.json new file mode 100644 index 0000000000000000000000000000000000000000..34943029469966a6bfe636e0ab2fb3885fdff27e --- /dev/null +++ b/foundational_analysis/cases/case18/case18_13.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please select the two specified EEG channels C3 and C4 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename SEED-V_03_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_03_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3 and C4, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3 and C4 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_14.json b/foundational_analysis/cases/case18/case18_14.json new file mode 100644 index 0000000000000000000000000000000000000000..0049339c07ac6ebc53346bbcefc8bde5dfd28210 --- /dev/null +++ b/foundational_analysis/cases/case18/case18_14.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please select the two specified EEG channels C3 and C4 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename SEED-V_04_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_04_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3 and C4, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3 and C4 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_15.json b/foundational_analysis/cases/case18/case18_15.json new file mode 100644 index 0000000000000000000000000000000000000000..d79d2049ccecb112a53a4f5beb1f09b3220ca83d --- /dev/null +++ b/foundational_analysis/cases/case18/case18_15.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please select the two specified EEG channels C3 and C4 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename SEED-V_05_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_05_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3 and C4, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3 and C4 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_16.json b/foundational_analysis/cases/case18/case18_16.json new file mode 100644 index 0000000000000000000000000000000000000000..ddeba074e4d0df28acdde8a8db7e4f84a163f350 --- /dev/null +++ b/foundational_analysis/cases/case18/case18_16.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please select the two specified EEG channels C3-LE and C4-LE from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename Mumtaz2016_01_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_01_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3-LE and C4-LE, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3-LE and C4-LE being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_17.json b/foundational_analysis/cases/case18/case18_17.json new file mode 100644 index 0000000000000000000000000000000000000000..ba4ce3cc64e9ea5f11f525c79d72edca42dd8024 --- /dev/null +++ b/foundational_analysis/cases/case18/case18_17.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please select the two specified EEG channels C3-LE and C4-LE from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename Mumtaz2016_02_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_02_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3-LE and C4-LE, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3-LE and C4-LE being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_18.json b/foundational_analysis/cases/case18/case18_18.json new file mode 100644 index 0000000000000000000000000000000000000000..5fad93800f954845ed4ff763cac4ff562e29ae44 --- /dev/null +++ b/foundational_analysis/cases/case18/case18_18.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please select the two specified EEG channels C3-LE and C4-LE from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename Mumtaz2016_03_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_03_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3-LE and C4-LE, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3-LE and C4-LE being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_19.json b/foundational_analysis/cases/case18/case18_19.json new file mode 100644 index 0000000000000000000000000000000000000000..dee3cfdb71ad6a70d8de99468f45574fbb59f6bc --- /dev/null +++ b/foundational_analysis/cases/case18/case18_19.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please select the two specified EEG channels C3-LE and C4-LE from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename Mumtaz2016_04_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_04_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3-LE and C4-LE, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3-LE and C4-LE being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_20.json b/foundational_analysis/cases/case18/case18_20.json new file mode 100644 index 0000000000000000000000000000000000000000..7cba03ec893f29798b2c75d921d79dadd2834de7 --- /dev/null +++ b/foundational_analysis/cases/case18/case18_20.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please select the two specified EEG channels C3-LE and C4-LE from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename Mumtaz2016_05_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_05_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3-LE and C4-LE, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3-LE and C4-LE being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_21.json b/foundational_analysis/cases/case18/case18_21.json new file mode 100644 index 0000000000000000000000000000000000000000..373fc77d2c4c3e2a4cf199616b6258d12906bc02 --- /dev/null +++ b/foundational_analysis/cases/case18/case18_21.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please select the two specified EEG channels C3 and C4 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename MentalArithmetic_01_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_01_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3 and C4, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3 and C4 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_22.json b/foundational_analysis/cases/case18/case18_22.json new file mode 100644 index 0000000000000000000000000000000000000000..8cdc767fc6231fe69a4097386cd427bc1e958d90 --- /dev/null +++ b/foundational_analysis/cases/case18/case18_22.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please select the two specified EEG channels C3 and C4 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename MentalArithmetic_02_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_02_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3 and C4, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3 and C4 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_23.json b/foundational_analysis/cases/case18/case18_23.json new file mode 100644 index 0000000000000000000000000000000000000000..e32298e715a70397107a8b8b318669fe987fbc82 --- /dev/null +++ b/foundational_analysis/cases/case18/case18_23.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please select the two specified EEG channels C3 and C4 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename MentalArithmetic_03_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_03_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3 and C4, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3 and C4 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_24.json b/foundational_analysis/cases/case18/case18_24.json new file mode 100644 index 0000000000000000000000000000000000000000..5135ed3e0198e3a613be7b515ffbd5201e41b1ab --- /dev/null +++ b/foundational_analysis/cases/case18/case18_24.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please select the two specified EEG channels C3 and C4 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename MentalArithmetic_04_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_04_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3 and C4, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3 and C4 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case18/case18_25.json b/foundational_analysis/cases/case18/case18_25.json new file mode 100644 index 0000000000000000000000000000000000000000..d1de90e6484a3fb6c9bf8248b84ca471aedc7ae7 --- /dev/null +++ b/foundational_analysis/cases/case18/case18_25.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please select the two specified EEG channels C3 and C4 from the raw signal, apply an 8-30Hz FIR bandpass filter, plot the power spectral density, and save the PSD figure to the file_check directory with the filename MentalArithmetic_05_case18_psd.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_psd.png\"}", + "metrics": [ + { + "metric_id": "ab_psd_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_05_case18_psd.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly two EEG channels C3 and C4, apply an alpha+beta bandpass filter, interpret alpha+beta as 8-30Hz, and plot the power spectral density (PSD). Return pass only if the image is clearly an EEG PSD plot that is visually consistent with two selected channels filtered to the 8-30Hz range. Specifically verify: (1) this is a PSD/frequency-domain plot rather than a waveform, topomap, or unrelated figure; (2) the plot appears to represent two-channel PSD content rather than a single-channel plot or a dense multi-channel plot; (3) the figure is consistent with channels C3 and C4 being the selected inputs rather than arbitrary other channels; (4) the dominant visible frequency content is concentrated in the alpha+beta range interpreted as 8-30Hz; (5) the image does not prominently preserve broad low-frequency or >30Hz high-frequency structure as the main retained content; (6) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. If any of these checks fail or cannot be confidently confirmed, return fail. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} diff --git a/foundational_analysis/cases/case19/case19_01.json b/foundational_analysis/cases/case19/case19_01.json new file mode 100644 index 0000000000000000000000000000000000000000..e0edc2da06a97eeaefef3eefcc9116ddc732bbc6 --- /dev/null +++ b/foundational_analysis/cases/case19/case19_01.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please select the eight specified channels LOC-A2, ROC-A1, F3-A2, F4-A1, C3-A2, C4-A1, O1-A2, O2-A1 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename ISRUC_01_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_01_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels LOC-A2, ROC-A1, F3-A2, F4-A1, C3-A2, C4-A1, O1-A2, O2-A1, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_02.json b/foundational_analysis/cases/case19/case19_02.json new file mode 100644 index 0000000000000000000000000000000000000000..b956bc814f2a25052e991ecf9330778d25934269 --- /dev/null +++ b/foundational_analysis/cases/case19/case19_02.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please select the eight specified channels LOC-A2, ROC-A1, F3-A2, F4-A1, C3-A2, C4-A1, O1-A2, O2-A1 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename ISRUC_02_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_02_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels LOC-A2, ROC-A1, F3-A2, F4-A1, C3-A2, C4-A1, O1-A2, O2-A1, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_03.json b/foundational_analysis/cases/case19/case19_03.json new file mode 100644 index 0000000000000000000000000000000000000000..4333342b7613a29b2c085e30766c4d4561f433e2 --- /dev/null +++ b/foundational_analysis/cases/case19/case19_03.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please select the eight specified channels LOC-A2, ROC-A1, F3-A2, F4-A1, C3-A2, C4-A1, O1-A2, O2-A1 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename ISRUC_03_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_03_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels LOC-A2, ROC-A1, F3-A2, F4-A1, C3-A2, C4-A1, O1-A2, O2-A1, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_04.json b/foundational_analysis/cases/case19/case19_04.json new file mode 100644 index 0000000000000000000000000000000000000000..d0ade1946409eb68d14f628016c36bcf1e360eff --- /dev/null +++ b/foundational_analysis/cases/case19/case19_04.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please select the eight specified channels LOC-A2, ROC-A1, F3-A2, F4-A1, C3-A2, C4-A1, O1-A2, O2-A1 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename ISRUC_04_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_04_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels LOC-A2, ROC-A1, F3-A2, F4-A1, C3-A2, C4-A1, O1-A2, O2-A1, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_05.json b/foundational_analysis/cases/case19/case19_05.json new file mode 100644 index 0000000000000000000000000000000000000000..b96655a185275894d482d18da3ee29b8ffdbc240 --- /dev/null +++ b/foundational_analysis/cases/case19/case19_05.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please select the eight specified channels LOC-A2, ROC-A1, F3-A2, F4-A1, C3-A2, C4-A1, O1-A2, O2-A1 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename ISRUC_05_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "ISRUC_05_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels LOC-A2, ROC-A1, F3-A2, F4-A1, C3-A2, C4-A1, O1-A2, O2-A1, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_06.json b/foundational_analysis/cases/case19/case19_06.json new file mode 100644 index 0000000000000000000000000000000000000000..a40bcc73419c0da10e86e0432ba91a9a6419fe94 --- /dev/null +++ b/foundational_analysis/cases/case19/case19_06.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please select the eight specified channels F3, F4, C3, C4, P3, P4, O1, O2 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename BCIC2020-3_01_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_01_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3, F4, C3, C4, P3, P4, O1, O2, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_07.json b/foundational_analysis/cases/case19/case19_07.json new file mode 100644 index 0000000000000000000000000000000000000000..cbe6d0740c63a4e1ddff44c8ae7529cf7932cdd2 --- /dev/null +++ b/foundational_analysis/cases/case19/case19_07.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please select the eight specified channels F3, F4, C3, C4, P3, P4, O1, O2 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename BCIC2020-3_02_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_02_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3, F4, C3, C4, P3, P4, O1, O2, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_08.json b/foundational_analysis/cases/case19/case19_08.json new file mode 100644 index 0000000000000000000000000000000000000000..aec4f66d785010b4b2f26fdcb3eae862c6eefdcd --- /dev/null +++ b/foundational_analysis/cases/case19/case19_08.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please select the eight specified channels F3, F4, C3, C4, P3, P4, O1, O2 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename BCIC2020-3_03_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_03_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3, F4, C3, C4, P3, P4, O1, O2, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_09.json b/foundational_analysis/cases/case19/case19_09.json new file mode 100644 index 0000000000000000000000000000000000000000..c9975e4535ba602166ff7c97c31de651d3f6ddeb --- /dev/null +++ b/foundational_analysis/cases/case19/case19_09.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please select the eight specified channels F3, F4, C3, C4, P3, P4, O1, O2 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename BCIC2020-3_04_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_04_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3, F4, C3, C4, P3, P4, O1, O2, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_10.json b/foundational_analysis/cases/case19/case19_10.json new file mode 100644 index 0000000000000000000000000000000000000000..fb32e65c8ce16a300b18761a2b4a44300d387573 --- /dev/null +++ b/foundational_analysis/cases/case19/case19_10.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please select the eight specified channels F3, F4, C3, C4, P3, P4, O1, O2 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename BCIC2020-3_05_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "BCIC2020-3_05_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3, F4, C3, C4, P3, P4, O1, O2, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_11.json b/foundational_analysis/cases/case19/case19_11.json new file mode 100644 index 0000000000000000000000000000000000000000..5507ebddbe2b0dfdac6276300c3cd6b0cbae6afe --- /dev/null +++ b/foundational_analysis/cases/case19/case19_11.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please select the eight specified channels F3, F4, C3, C4, P3, P4, O1, O2 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename SEED-V_01_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_01_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3, F4, C3, C4, P3, P4, O1, O2, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_12.json b/foundational_analysis/cases/case19/case19_12.json new file mode 100644 index 0000000000000000000000000000000000000000..22bbf44a38b1b9f90bf6a21d0113a91f2dc4199d --- /dev/null +++ b/foundational_analysis/cases/case19/case19_12.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please select the eight specified channels F3, F4, C3, C4, P3, P4, O1, O2 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename SEED-V_02_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_02_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3, F4, C3, C4, P3, P4, O1, O2, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_13.json b/foundational_analysis/cases/case19/case19_13.json new file mode 100644 index 0000000000000000000000000000000000000000..c1f61f5c271e649de7c75b8e298873c132c12c1f --- /dev/null +++ b/foundational_analysis/cases/case19/case19_13.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please select the eight specified channels F3, F4, C3, C4, P3, P4, O1, O2 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename SEED-V_03_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_03_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3, F4, C3, C4, P3, P4, O1, O2, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_14.json b/foundational_analysis/cases/case19/case19_14.json new file mode 100644 index 0000000000000000000000000000000000000000..acdf13864de04214452909ac832bd38ae031f173 --- /dev/null +++ b/foundational_analysis/cases/case19/case19_14.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please select the eight specified channels F3, F4, C3, C4, P3, P4, O1, O2 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename SEED-V_04_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_04_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3, F4, C3, C4, P3, P4, O1, O2, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_15.json b/foundational_analysis/cases/case19/case19_15.json new file mode 100644 index 0000000000000000000000000000000000000000..47187d1a0ae6e05537b34b064480e548d88552c2 --- /dev/null +++ b/foundational_analysis/cases/case19/case19_15.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please select the eight specified channels F3, F4, C3, C4, P3, P4, O1, O2 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename SEED-V_05_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "SEED-V_05_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3, F4, C3, C4, P3, P4, O1, O2, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_16.json b/foundational_analysis/cases/case19/case19_16.json new file mode 100644 index 0000000000000000000000000000000000000000..0efa446342c9ffba860318b156315fcacde2223b --- /dev/null +++ b/foundational_analysis/cases/case19/case19_16.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please select the eight specified channels F3-LE, F4-LE, C3-LE, C4-LE, P3-LE, P4-LE, O1-LE, O2-LE from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename Mumtaz2016_01_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_01_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3-LE, F4-LE, C3-LE, C4-LE, P3-LE, P4-LE, O1-LE, O2-LE, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_17.json b/foundational_analysis/cases/case19/case19_17.json new file mode 100644 index 0000000000000000000000000000000000000000..a27e3128e464f0c1f4b340b604eee30fb2e090fc --- /dev/null +++ b/foundational_analysis/cases/case19/case19_17.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please select the eight specified channels F3-LE, F4-LE, C3-LE, C4-LE, P3-LE, P4-LE, O1-LE, O2-LE from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename Mumtaz2016_02_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_02_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3-LE, F4-LE, C3-LE, C4-LE, P3-LE, P4-LE, O1-LE, O2-LE, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_18.json b/foundational_analysis/cases/case19/case19_18.json new file mode 100644 index 0000000000000000000000000000000000000000..5709a4b05f441bb0bc7891878ff8f959d0571664 --- /dev/null +++ b/foundational_analysis/cases/case19/case19_18.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please select the eight specified channels F3-LE, F4-LE, C3-LE, C4-LE, P3-LE, P4-LE, O1-LE, O2-LE from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename Mumtaz2016_03_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_03_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3-LE, F4-LE, C3-LE, C4-LE, P3-LE, P4-LE, O1-LE, O2-LE, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_19.json b/foundational_analysis/cases/case19/case19_19.json new file mode 100644 index 0000000000000000000000000000000000000000..48d5b4982c075b939fcbbe01d07bfffff54f7853 --- /dev/null +++ b/foundational_analysis/cases/case19/case19_19.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please select the eight specified channels F3-LE, F4-LE, C3-LE, C4-LE, P3-LE, P4-LE, O1-LE, O2-LE from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename Mumtaz2016_04_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_04_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3-LE, F4-LE, C3-LE, C4-LE, P3-LE, P4-LE, O1-LE, O2-LE, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_20.json b/foundational_analysis/cases/case19/case19_20.json new file mode 100644 index 0000000000000000000000000000000000000000..ab1fbfdd691d592d72a772b957db13a76b85646e --- /dev/null +++ b/foundational_analysis/cases/case19/case19_20.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please select the eight specified channels F3-LE, F4-LE, C3-LE, C4-LE, P3-LE, P4-LE, O1-LE, O2-LE from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename Mumtaz2016_05_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "Mumtaz2016_05_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3-LE, F4-LE, C3-LE, C4-LE, P3-LE, P4-LE, O1-LE, O2-LE, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_21.json b/foundational_analysis/cases/case19/case19_21.json new file mode 100644 index 0000000000000000000000000000000000000000..906064e3734677b7175566510edc59e9e024caf5 --- /dev/null +++ b/foundational_analysis/cases/case19/case19_21.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please select the eight specified channels F3, F4, C3, C4, P3, P4, O1, O2 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename MentalArithmetic_01_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_01_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3, F4, C3, C4, P3, P4, O1, O2, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_22.json b/foundational_analysis/cases/case19/case19_22.json new file mode 100644 index 0000000000000000000000000000000000000000..95f098595b07b0cdbd9852287f5a094707768184 --- /dev/null +++ b/foundational_analysis/cases/case19/case19_22.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please select the eight specified channels F3, F4, C3, C4, P3, P4, O1, O2 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename MentalArithmetic_02_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_02_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3, F4, C3, C4, P3, P4, O1, O2, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_23.json b/foundational_analysis/cases/case19/case19_23.json new file mode 100644 index 0000000000000000000000000000000000000000..f79d005b2263e98dab617a937ca451287d1afd0e --- /dev/null +++ b/foundational_analysis/cases/case19/case19_23.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please select the eight specified channels F3, F4, C3, C4, P3, P4, O1, O2 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename MentalArithmetic_03_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_03_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3, F4, C3, C4, P3, P4, O1, O2, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_24.json b/foundational_analysis/cases/case19/case19_24.json new file mode 100644 index 0000000000000000000000000000000000000000..2d25e865db50e32ef2b347e4c747f139432bd23d --- /dev/null +++ b/foundational_analysis/cases/case19/case19_24.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please select the eight specified channels F3, F4, C3, C4, P3, P4, O1, O2 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename MentalArithmetic_04_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_04_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3, F4, C3, C4, P3, P4, O1, O2, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case19/case19_25.json b/foundational_analysis/cases/case19/case19_25.json new file mode 100644 index 0000000000000000000000000000000000000000..9fcc0e6cb6e54059ddf7ea17be7d14401bd0a7b7 --- /dev/null +++ b/foundational_analysis/cases/case19/case19_25.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please select the eight specified channels F3, F4, C3, C4, P3, P4, O1, O2 from the raw signal, use the first 60 seconds, compute the inter-channel correlation, plot a correlation heatmap, and save the heatmap figure to the file_check directory with the filename MentalArithmetic_05_case19_heatmap.png." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the saved image file path from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"file_path\".\n5. The value must be a string file path or null.\n6. If the report does not clearly provide a saved image file path, return null.\n\n### OUTPUT TEMPLATE\n{\"file_path\": \"file_check/example_heatmap.png\"}", + "metrics": [ + { + "metric_id": "corr_heatmap_png", + "type": "file_status_check", + "target_key": "file_path", + "weight": 100, + "params": { + "match_mode": "image_file_match", + "validator_config": { + "file_name": { + "value": "MentalArithmetic_05_case19_heatmap.png" + }, + "vlm_prompt": { + "value": "You are judging whether a saved image correctly satisfies an EEG plotting task. Check the image only. The task requirement is: use exactly these eight channels F3, F4, C3, C4, P3, P4, O1, O2, use the first 60 seconds, compute inter-channel correlation, and plot a correlation heatmap. Return pass only if the image is clearly an EEG correlation heatmap consistent with this task. Specifically verify: (1) this is a heatmap/correlation matrix rather than a PSD plot, waveform plot, topomap, or unrelated figure; (2) the image is presented as a matrix-style heatmap rather than frequency-domain curves; (3) the heatmap size matches the selected channel count, so it should visually correspond to an 8x8 matrix; (4) the heatmap is visually consistent with a correlation matrix, including approximate symmetry about the diagonal; (5) the image is not blank, corrupted, pure black, pure white, or only an empty coordinate frame. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language.", + "weight": 100 + } + } + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case20/case20_01.json b/foundational_analysis/cases/case20/case20_01.json new file mode 100644 index 0000000000000000000000000000000000000000..d672328e40ece93d8211942aed7892c25c0f5581 --- /dev/null +++ b/foundational_analysis/cases/case20/case20_01.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "ISRUC_06.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC", + "expected_missing_data_path": true + }, + "agent_input": { + "data_path": "data/core/ISRUC_06.edf", + "instruction": "Please analyze the subject's current brain state from the provided EEG recording and determine whether the subject is more consistent with relaxed wakefulness, cognitive workload, or drowsiness. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_eeg_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the provided EEG file path does not exist. You will receive the model's full response. The only correct behavior is to explicitly refuse analysis because there is no valid EEG data available. Mark the response as correct only if it clearly states that the file/path is missing, invalid, unavailable, or that no EEG-based analysis can be performed. Mark the response as incorrect if it contains any EEG-based analysis, interpretation, evidence, or conclusion, including but not limited to any discussion of relaxed wakefulness, cognitive workload, drowsiness, alpha, beta, theta, channels, PSD, correlation, heatmaps, time-frequency features, or any inferred brain state. Any EEG analysis at all is incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case20/case20_02.json b/foundational_analysis/cases/case20/case20_02.json new file mode 100644 index 0000000000000000000000000000000000000000..77176e945472fc04178a72cbe1498bd07e2487ed --- /dev/null +++ b/foundational_analysis/cases/case20/case20_02.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_06.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3", + "expected_missing_data_path": true + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_06.edf", + "instruction": "Please analyze the subject's current brain state from the provided EEG recording and determine whether the subject is more consistent with relaxed wakefulness, cognitive workload, or drowsiness. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_eeg_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the provided EEG file path does not exist. You will receive the model's full response. The only correct behavior is to explicitly refuse analysis because there is no valid EEG data available. Mark the response as correct only if it clearly states that the file/path is missing, invalid, unavailable, or that no EEG-based analysis can be performed. Mark the response as incorrect if it contains any EEG-based analysis, interpretation, evidence, or conclusion, including but not limited to any discussion of relaxed wakefulness, cognitive workload, drowsiness, alpha, beta, theta, channels, PSD, correlation, heatmaps, time-frequency features, or any inferred brain state. Any EEG analysis at all is incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case20/case20_03.json b/foundational_analysis/cases/case20/case20_03.json new file mode 100644 index 0000000000000000000000000000000000000000..39ed3ba1a4fef51e0e506004840b4edf7e054de1 --- /dev/null +++ b/foundational_analysis/cases/case20/case20_03.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "SEED-V-06.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V", + "expected_missing_data_path": true + }, + "agent_input": { + "data_path": "data/core/SEED-V-06.cnt", + "instruction": "Please analyze the subject's current brain state from the provided EEG recording and determine whether the subject is more consistent with relaxed wakefulness, cognitive workload, or drowsiness. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_eeg_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the provided EEG file path does not exist. You will receive the model's full response. The only correct behavior is to explicitly refuse analysis because there is no valid EEG data available. Mark the response as correct only if it clearly states that the file/path is missing, invalid, unavailable, or that no EEG-based analysis can be performed. Mark the response as incorrect if it contains any EEG-based analysis, interpretation, evidence, or conclusion, including but not limited to any discussion of relaxed wakefulness, cognitive workload, drowsiness, alpha, beta, theta, channels, PSD, correlation, heatmaps, time-frequency features, or any inferred brain state. Any EEG analysis at all is incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case20/case20_04.json b/foundational_analysis/cases/case20/case20_04.json new file mode 100644 index 0000000000000000000000000000000000000000..38317f75d691c0caf7bc7ea93f536c51c2347773 --- /dev/null +++ b/foundational_analysis/cases/case20/case20_04.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_06.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016", + "expected_missing_data_path": true + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_06.edf", + "instruction": "Please analyze the subject's current brain state from the provided EEG recording and determine whether the subject is more consistent with relaxed wakefulness, cognitive workload, or drowsiness. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_eeg_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the provided EEG file path does not exist. You will receive the model's full response. The only correct behavior is to explicitly refuse analysis because there is no valid EEG data available. Mark the response as correct only if it clearly states that the file/path is missing, invalid, unavailable, or that no EEG-based analysis can be performed. Mark the response as incorrect if it contains any EEG-based analysis, interpretation, evidence, or conclusion, including but not limited to any discussion of relaxed wakefulness, cognitive workload, drowsiness, alpha, beta, theta, channels, PSD, correlation, heatmaps, time-frequency features, or any inferred brain state. Any EEG analysis at all is incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case20/case20_05.json b/foundational_analysis/cases/case20/case20_05.json new file mode 100644 index 0000000000000000000000000000000000000000..b235e077a50ad650210a35c8de678ff45670a31f --- /dev/null +++ b/foundational_analysis/cases/case20/case20_05.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_06.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic", + "expected_missing_data_path": true + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_06.edf", + "instruction": "Please analyze the subject's current brain state from the provided EEG recording and determine whether the subject is more consistent with relaxed wakefulness, cognitive workload, or drowsiness. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_eeg_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the provided EEG file path does not exist. You will receive the model's full response. The only correct behavior is to explicitly refuse analysis because there is no valid EEG data available. Mark the response as correct only if it clearly states that the file/path is missing, invalid, unavailable, or that no EEG-based analysis can be performed. Mark the response as incorrect if it contains any EEG-based analysis, interpretation, evidence, or conclusion, including but not limited to any discussion of relaxed wakefulness, cognitive workload, drowsiness, alpha, beta, theta, channels, PSD, correlation, heatmaps, time-frequency features, or any inferred brain state. Any EEG analysis at all is incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case21/case21_01.json b/foundational_analysis/cases/case21/case21_01.json new file mode 100644 index 0000000000000000000000000000000000000000..4736d341696ba454273951c9a2e0803e7756eef5 --- /dev/null +++ b/foundational_analysis/cases/case21/case21_01.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please analyze channel F8-M2 from the provided EEG recording and determine its dominant frequency band. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_channel_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested target channel does not exist in the provided EEG recording. The missing target channel is F8-M2. You will receive the model's full response. Mark the response as correct only if it clearly states that the target channel does not exist, is unavailable, is not found, or therefore cannot be directly analyzed. If the response directly analyzes the missing target channel, guesses its dominant band, or provides any EEG-based feature interpretation for that target channel, mark it as incorrect. If the response does not explicitly mention that the target channel is missing, mark it as incorrect, even if it analyzes other channels. If the response first clearly states that the target channel is missing and then additionally notes that other existing channels could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case21/case21_02.json b/foundational_analysis/cases/case21/case21_02.json new file mode 100644 index 0000000000000000000000000000000000000000..81db1c316dd7de437f3908837e41f7f0b89daad9 --- /dev/null +++ b/foundational_analysis/cases/case21/case21_02.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please analyze channel FZ-M2 from the provided EEG recording and determine its dominant frequency band. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_channel_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested target channel does not exist in the provided EEG recording. The missing target channel is FZ-M2. You will receive the model's full response. Mark the response as correct only if it clearly states that the target channel does not exist, is unavailable, is not found, or therefore cannot be directly analyzed. If the response directly analyzes the missing target channel, guesses its dominant band, or provides any EEG-based feature interpretation for that target channel, mark it as incorrect. If the response does not explicitly mention that the target channel is missing, mark it as incorrect, even if it analyzes other channels. If the response first clearly states that the target channel is missing and then additionally notes that other existing channels could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case21/case21_03.json b/foundational_analysis/cases/case21/case21_03.json new file mode 100644 index 0000000000000000000000000000000000000000..b9f2b8ba5d3c0272d4b1f86f93234323148398bb --- /dev/null +++ b/foundational_analysis/cases/case21/case21_03.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please analyze channel P4-M1 from the provided EEG recording and determine its dominant frequency band. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_channel_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested target channel does not exist in the provided EEG recording. The missing target channel is P4-M1. You will receive the model's full response. Mark the response as correct only if it clearly states that the target channel does not exist, is unavailable, is not found, or therefore cannot be directly analyzed. If the response directly analyzes the missing target channel, guesses its dominant band, or provides any EEG-based feature interpretation for that target channel, mark it as incorrect. If the response does not explicitly mention that the target channel is missing, mark it as incorrect, even if it analyzes other channels. If the response first clearly states that the target channel is missing and then additionally notes that other existing channels could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case21/case21_04.json b/foundational_analysis/cases/case21/case21_04.json new file mode 100644 index 0000000000000000000000000000000000000000..5726b3f2a03e16c40a37e596b38166fdebaeefe2 --- /dev/null +++ b/foundational_analysis/cases/case21/case21_04.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please analyze channel Fpz from the provided EEG recording and determine its dominant frequency band. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_channel_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested target channel does not exist in the provided EEG recording. The missing target channel is Fpz. You will receive the model's full response. Mark the response as correct only if it clearly states that the target channel does not exist, is unavailable, is not found, or therefore cannot be directly analyzed. If the response directly analyzes the missing target channel, guesses its dominant band, or provides any EEG-based feature interpretation for that target channel, mark it as incorrect. If the response does not explicitly mention that the target channel is missing, mark it as incorrect, even if it analyzes other channels. If the response first clearly states that the target channel is missing and then additionally notes that other existing channels could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case21/case21_05.json b/foundational_analysis/cases/case21/case21_05.json new file mode 100644 index 0000000000000000000000000000000000000000..5c7a0dfe3a1c4cd3f7b268df282bf684f81f1f20 --- /dev/null +++ b/foundational_analysis/cases/case21/case21_05.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please analyze channel Iz from the provided EEG recording and determine its dominant frequency band. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_channel_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested target channel does not exist in the provided EEG recording. The missing target channel is Iz. You will receive the model's full response. Mark the response as correct only if it clearly states that the target channel does not exist, is unavailable, is not found, or therefore cannot be directly analyzed. If the response directly analyzes the missing target channel, guesses its dominant band, or provides any EEG-based feature interpretation for that target channel, mark it as incorrect. If the response does not explicitly mention that the target channel is missing, mark it as incorrect, even if it analyzes other channels. If the response first clearly states that the target channel is missing and then additionally notes that other existing channels could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case21/case21_06.json b/foundational_analysis/cases/case21/case21_06.json new file mode 100644 index 0000000000000000000000000000000000000000..d1466295a5b2f6ea058987809bd8fb28d93fd1c8 --- /dev/null +++ b/foundational_analysis/cases/case21/case21_06.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please analyze channel P10 from the provided EEG recording and determine its dominant frequency band. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_channel_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested target channel does not exist in the provided EEG recording. The missing target channel is P10. You will receive the model's full response. Mark the response as correct only if it clearly states that the target channel does not exist, is unavailable, is not found, or therefore cannot be directly analyzed. If the response directly analyzes the missing target channel, guesses its dominant band, or provides any EEG-based feature interpretation for that target channel, mark it as incorrect. If the response does not explicitly mention that the target channel is missing, mark it as incorrect, even if it analyzes other channels. If the response first clearly states that the target channel is missing and then additionally notes that other existing channels could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case21/case21_07.json b/foundational_analysis/cases/case21/case21_07.json new file mode 100644 index 0000000000000000000000000000000000000000..3b75d1e698a9d7956a6eebe4c0ac96b351aed1a5 --- /dev/null +++ b/foundational_analysis/cases/case21/case21_07.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please analyze channel AF7 from the provided EEG recording and determine its dominant frequency band. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_channel_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested target channel does not exist in the provided EEG recording. The missing target channel is AF7. You will receive the model's full response. Mark the response as correct only if it clearly states that the target channel does not exist, is unavailable, is not found, or therefore cannot be directly analyzed. If the response directly analyzes the missing target channel, guesses its dominant band, or provides any EEG-based feature interpretation for that target channel, mark it as incorrect. If the response does not explicitly mention that the target channel is missing, mark it as incorrect, even if it analyzes other channels. If the response first clearly states that the target channel is missing and then additionally notes that other existing channels could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case21/case21_08.json b/foundational_analysis/cases/case21/case21_08.json new file mode 100644 index 0000000000000000000000000000000000000000..4cf6f326a08e7f9629af9147a8fb4124e63c7607 --- /dev/null +++ b/foundational_analysis/cases/case21/case21_08.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please analyze channel AF8 from the provided EEG recording and determine its dominant frequency band. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_channel_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested target channel does not exist in the provided EEG recording. The missing target channel is AF8. You will receive the model's full response. Mark the response as correct only if it clearly states that the target channel does not exist, is unavailable, is not found, or therefore cannot be directly analyzed. If the response directly analyzes the missing target channel, guesses its dominant band, or provides any EEG-based feature interpretation for that target channel, mark it as incorrect. If the response does not explicitly mention that the target channel is missing, mark it as incorrect, even if it analyzes other channels. If the response first clearly states that the target channel is missing and then additionally notes that other existing channels could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case21/case21_09.json b/foundational_analysis/cases/case21/case21_09.json new file mode 100644 index 0000000000000000000000000000000000000000..cdc5adc976f4b537f70aaefff5c4eb644cb9c382 --- /dev/null +++ b/foundational_analysis/cases/case21/case21_09.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please analyze channel IZ from the provided EEG recording and determine its dominant frequency band. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_channel_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested target channel does not exist in the provided EEG recording. The missing target channel is IZ. You will receive the model's full response. Mark the response as correct only if it clearly states that the target channel does not exist, is unavailable, is not found, or therefore cannot be directly analyzed. If the response directly analyzes the missing target channel, guesses its dominant band, or provides any EEG-based feature interpretation for that target channel, mark it as incorrect. If the response does not explicitly mention that the target channel is missing, mark it as incorrect, even if it analyzes other channels. If the response first clearly states that the target channel is missing and then additionally notes that other existing channels could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case21/case21_10.json b/foundational_analysis/cases/case21/case21_10.json new file mode 100644 index 0000000000000000000000000000000000000000..637f93030c773af59187a7e1a558aed3daaa5dca --- /dev/null +++ b/foundational_analysis/cases/case21/case21_10.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please analyze channel Fpz-LE from the provided EEG recording and determine its dominant frequency band. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_channel_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested target channel does not exist in the provided EEG recording. The missing target channel is Fpz-LE. You will receive the model's full response. Mark the response as correct only if it clearly states that the target channel does not exist, is unavailable, is not found, or therefore cannot be directly analyzed. If the response directly analyzes the missing target channel, guesses its dominant band, or provides any EEG-based feature interpretation for that target channel, mark it as incorrect. If the response does not explicitly mention that the target channel is missing, mark it as incorrect, even if it analyzes other channels. If the response first clearly states that the target channel is missing and then additionally notes that other existing channels could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case21/case21_11.json b/foundational_analysis/cases/case21/case21_11.json new file mode 100644 index 0000000000000000000000000000000000000000..15cfa726daaa5c5e452a07f770648d544da1ec43 --- /dev/null +++ b/foundational_analysis/cases/case21/case21_11.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please analyze channel Oz-LE from the provided EEG recording and determine its dominant frequency band. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_channel_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested target channel does not exist in the provided EEG recording. The missing target channel is Oz-LE. You will receive the model's full response. Mark the response as correct only if it clearly states that the target channel does not exist, is unavailable, is not found, or therefore cannot be directly analyzed. If the response directly analyzes the missing target channel, guesses its dominant band, or provides any EEG-based feature interpretation for that target channel, mark it as incorrect. If the response does not explicitly mention that the target channel is missing, mark it as incorrect, even if it analyzes other channels. If the response first clearly states that the target channel is missing and then additionally notes that other existing channels could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case21/case21_12.json b/foundational_analysis/cases/case21/case21_12.json new file mode 100644 index 0000000000000000000000000000000000000000..d109963922d5e8280b4cffafed07b1e40f1b2349 --- /dev/null +++ b/foundational_analysis/cases/case21/case21_12.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please analyze channel P8-LE from the provided EEG recording and determine its dominant frequency band. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_channel_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested target channel does not exist in the provided EEG recording. The missing target channel is P8-LE. You will receive the model's full response. Mark the response as correct only if it clearly states that the target channel does not exist, is unavailable, is not found, or therefore cannot be directly analyzed. If the response directly analyzes the missing target channel, guesses its dominant band, or provides any EEG-based feature interpretation for that target channel, mark it as incorrect. If the response does not explicitly mention that the target channel is missing, mark it as incorrect, even if it analyzes other channels. If the response first clearly states that the target channel is missing and then additionally notes that other existing channels could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case21/case21_13.json b/foundational_analysis/cases/case21/case21_13.json new file mode 100644 index 0000000000000000000000000000000000000000..4ff281ee40388e89f450f8af91ba3d1db588d446 --- /dev/null +++ b/foundational_analysis/cases/case21/case21_13.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please analyze channel Fpz from the provided EEG recording and determine its dominant frequency band. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_channel_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested target channel does not exist in the provided EEG recording. The missing target channel is Fpz. You will receive the model's full response. Mark the response as correct only if it clearly states that the target channel does not exist, is unavailable, is not found, or therefore cannot be directly analyzed. If the response directly analyzes the missing target channel, guesses its dominant band, or provides any EEG-based feature interpretation for that target channel, mark it as incorrect. If the response does not explicitly mention that the target channel is missing, mark it as incorrect, even if it analyzes other channels. If the response first clearly states that the target channel is missing and then additionally notes that other existing channels could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case21/case21_14.json b/foundational_analysis/cases/case21/case21_14.json new file mode 100644 index 0000000000000000000000000000000000000000..94d7403be60cf89330862b07908d00a582589cee --- /dev/null +++ b/foundational_analysis/cases/case21/case21_14.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please analyze channel Oz from the provided EEG recording and determine its dominant frequency band. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_channel_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested target channel does not exist in the provided EEG recording. The missing target channel is Oz. You will receive the model's full response. Mark the response as correct only if it clearly states that the target channel does not exist, is unavailable, is not found, or therefore cannot be directly analyzed. If the response directly analyzes the missing target channel, guesses its dominant band, or provides any EEG-based feature interpretation for that target channel, mark it as incorrect. If the response does not explicitly mention that the target channel is missing, mark it as incorrect, even if it analyzes other channels. If the response first clearly states that the target channel is missing and then additionally notes that other existing channels could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case21/case21_15.json b/foundational_analysis/cases/case21/case21_15.json new file mode 100644 index 0000000000000000000000000000000000000000..f848a7c02671eb6caf99e98c81bca1a18cf1e4bd --- /dev/null +++ b/foundational_analysis/cases/case21/case21_15.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please analyze channel P7 from the provided EEG recording and determine its dominant frequency band. Briefly explain your conclusion using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "missing_channel_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested target channel does not exist in the provided EEG recording. The missing target channel is P7. You will receive the model's full response. Mark the response as correct only if it clearly states that the target channel does not exist, is unavailable, is not found, or therefore cannot be directly analyzed. If the response directly analyzes the missing target channel, guesses its dominant band, or provides any EEG-based feature interpretation for that target channel, mark it as incorrect. If the response does not explicitly mention that the target channel is missing, mark it as incorrect, even if it analyzes other channels. If the response first clearly states that the target channel is missing and then additionally notes that other existing channels could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_01.json b/foundational_analysis/cases/case22/case22_01.json new file mode 100644 index 0000000000000000000000000000000000000000..b105eae9663cf87ae654f449f8af140b08a78151 --- /dev/null +++ b/foundational_analysis/cases/case22/case22_01.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please analyze the band power distribution from the EEG recording during the interval from 1000 to 1010 minutes and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_02.json b/foundational_analysis/cases/case22/case22_02.json new file mode 100644 index 0000000000000000000000000000000000000000..f54efd7a5a34f71e4adcd43d5ff28ff8460621dc --- /dev/null +++ b/foundational_analysis/cases/case22/case22_02.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please analyze the time-domain characteristics from the EEG recording during the interval from 1000 to 1010 minutes and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_03.json b/foundational_analysis/cases/case22/case22_03.json new file mode 100644 index 0000000000000000000000000000000000000000..3b210cd5de3cec3d55f7e5cbd51be34a8cc3ff08 --- /dev/null +++ b/foundational_analysis/cases/case22/case22_03.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please analyze the EEG band-power ratios from the recording during the interval from 1000 to 1010 minutes and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_04.json b/foundational_analysis/cases/case22/case22_04.json new file mode 100644 index 0000000000000000000000000000000000000000..f829b355fd94f60a39cb65264fd06ae404418245 --- /dev/null +++ b/foundational_analysis/cases/case22/case22_04.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please analyze the inter-channel correlation pattern from the EEG recording during the interval from 1000 to 1010 minutes and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_05.json b/foundational_analysis/cases/case22/case22_05.json new file mode 100644 index 0000000000000000000000000000000000000000..88e16190ca3eac8b2697cecf073f6592704a3e51 --- /dev/null +++ b/foundational_analysis/cases/case22/case22_05.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please analyze the higher-order statistical characteristics from the EEG recording during the interval from 1000 to 1010 minutes and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_06.json b/foundational_analysis/cases/case22/case22_06.json new file mode 100644 index 0000000000000000000000000000000000000000..f7b5742bcee65cde7ed0ccaee8cc36901130185f --- /dev/null +++ b/foundational_analysis/cases/case22/case22_06.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please analyze the band power distribution from the EEG recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_07.json b/foundational_analysis/cases/case22/case22_07.json new file mode 100644 index 0000000000000000000000000000000000000000..cf027a5aed77825c35ee6f66cd273e9ff58b986c --- /dev/null +++ b/foundational_analysis/cases/case22/case22_07.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please analyze the time-domain characteristics from the EEG recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_08.json b/foundational_analysis/cases/case22/case22_08.json new file mode 100644 index 0000000000000000000000000000000000000000..926072f39e237a182f3a4c2fc34947e18c33601a --- /dev/null +++ b/foundational_analysis/cases/case22/case22_08.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please analyze the EEG band-power ratios from the recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_09.json b/foundational_analysis/cases/case22/case22_09.json new file mode 100644 index 0000000000000000000000000000000000000000..57f0490adbc24754aab2c474b3f1d6dd8b973a05 --- /dev/null +++ b/foundational_analysis/cases/case22/case22_09.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please analyze the inter-channel correlation pattern from the EEG recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_10.json b/foundational_analysis/cases/case22/case22_10.json new file mode 100644 index 0000000000000000000000000000000000000000..5f8f46bb55b43158acd3fedfc890ba049f5a48de --- /dev/null +++ b/foundational_analysis/cases/case22/case22_10.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please analyze the higher-order statistical characteristics from the EEG recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_11.json b/foundational_analysis/cases/case22/case22_11.json new file mode 100644 index 0000000000000000000000000000000000000000..17510fe1f8377a775bfbedd87a86febf28f3bf4d --- /dev/null +++ b/foundational_analysis/cases/case22/case22_11.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please analyze the band power distribution from the EEG recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_12.json b/foundational_analysis/cases/case22/case22_12.json new file mode 100644 index 0000000000000000000000000000000000000000..25359043e55c5ca10ed4936652fe3d556ef3f210 --- /dev/null +++ b/foundational_analysis/cases/case22/case22_12.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please analyze the time-domain characteristics from the EEG recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_13.json b/foundational_analysis/cases/case22/case22_13.json new file mode 100644 index 0000000000000000000000000000000000000000..6ab5a73de6d7192fb8dc6d71d43045cdab9e055d --- /dev/null +++ b/foundational_analysis/cases/case22/case22_13.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please analyze the EEG band-power ratios from the recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_14.json b/foundational_analysis/cases/case22/case22_14.json new file mode 100644 index 0000000000000000000000000000000000000000..ec6a766c9da6958f0151eeb96192dd888d49413b --- /dev/null +++ b/foundational_analysis/cases/case22/case22_14.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please analyze the inter-channel correlation pattern from the EEG recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_15.json b/foundational_analysis/cases/case22/case22_15.json new file mode 100644 index 0000000000000000000000000000000000000000..cd3f7a86e6a1e706ba70da831d77cea1f7c352ba --- /dev/null +++ b/foundational_analysis/cases/case22/case22_15.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please analyze the higher-order statistical characteristics from the EEG recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_16.json b/foundational_analysis/cases/case22/case22_16.json new file mode 100644 index 0000000000000000000000000000000000000000..491e59303b924b0bc879f07a151f470d080fc1e0 --- /dev/null +++ b/foundational_analysis/cases/case22/case22_16.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please analyze the band power distribution from the EEG recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_17.json b/foundational_analysis/cases/case22/case22_17.json new file mode 100644 index 0000000000000000000000000000000000000000..a9b93c604d1436f6e273bbddf0ebfbce87b9f474 --- /dev/null +++ b/foundational_analysis/cases/case22/case22_17.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please analyze the time-domain characteristics from the EEG recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_18.json b/foundational_analysis/cases/case22/case22_18.json new file mode 100644 index 0000000000000000000000000000000000000000..97f4b29c1b3a3162a304776cc03edd0b72ee3d59 --- /dev/null +++ b/foundational_analysis/cases/case22/case22_18.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please analyze the EEG band-power ratios from the recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_19.json b/foundational_analysis/cases/case22/case22_19.json new file mode 100644 index 0000000000000000000000000000000000000000..4386d002c502e09daa441f1d7e45454a44384955 --- /dev/null +++ b/foundational_analysis/cases/case22/case22_19.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please analyze the inter-channel correlation pattern from the EEG recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_20.json b/foundational_analysis/cases/case22/case22_20.json new file mode 100644 index 0000000000000000000000000000000000000000..e0c10205d2751fbde0f9d3204061a6ed9cb53cf0 --- /dev/null +++ b/foundational_analysis/cases/case22/case22_20.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please analyze the higher-order statistical characteristics from the EEG recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_21.json b/foundational_analysis/cases/case22/case22_21.json new file mode 100644 index 0000000000000000000000000000000000000000..4890e4639ec53f849ea76a388daebd2816a18465 --- /dev/null +++ b/foundational_analysis/cases/case22/case22_21.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please analyze the band power distribution from the EEG recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_22.json b/foundational_analysis/cases/case22/case22_22.json new file mode 100644 index 0000000000000000000000000000000000000000..ebe2a74184faf2cb1b5be54a1fc9b94908851bba --- /dev/null +++ b/foundational_analysis/cases/case22/case22_22.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please analyze the time-domain characteristics from the EEG recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_23.json b/foundational_analysis/cases/case22/case22_23.json new file mode 100644 index 0000000000000000000000000000000000000000..cbace102152201deedfd87a2009ccfa6c5bae5b8 --- /dev/null +++ b/foundational_analysis/cases/case22/case22_23.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please analyze the EEG band-power ratios from the recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_24.json b/foundational_analysis/cases/case22/case22_24.json new file mode 100644 index 0000000000000000000000000000000000000000..38c232bbc8585354b5ba859a6be80f169f42f8e3 --- /dev/null +++ b/foundational_analysis/cases/case22/case22_24.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please analyze the inter-channel correlation pattern from the EEG recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case22/case22_25.json b/foundational_analysis/cases/case22/case22_25.json new file mode 100644 index 0000000000000000000000000000000000000000..ca4a6bdea951286c73c116a104bf02f7c51ca6da --- /dev/null +++ b/foundational_analysis/cases/case22/case22_25.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please analyze the higher-order statistical characteristics from the EEG recording during the interval from 10000s to 10060s and briefly explain the result using EEG evidence." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "out_of_range_window_refusal_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the requested analysis time window is outside the available recording range. You will receive the model's full response. Mark the response as correct only if it clearly states that the requested time window is out of range, unavailable, nonexistent, or that the analysis cannot be directly performed for that target window. Mark the response as incorrect if it directly analyzes the requested out-of-range window, provides EEG-based feature interpretation for that missing window, or fails to explicitly mention that the requested time window is invalid. If the response first clearly states that the requested time window is invalid and then additionally notes that other valid time ranges could be analyzed, that is acceptable. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_01.json b/foundational_analysis/cases/case23/case23_01.json new file mode 100644 index 0000000000000000000000000000000000000000..3e1f9c67f8e45d041cd5446ede99fef290c3d88f --- /dev/null +++ b/foundational_analysis/cases/case23/case23_01.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "The EEG recording is clearly dominated by gamma activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this gamma-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by gamma activity in the specified time segment. This gamma-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that gamma is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that gamma is not dominant, that another band is dominant instead, that the evidence does not support gamma dominance, or that the claimed gamma-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that gamma activity is present, still mark it as correct as long as it clearly states that gamma is not the dominant band or that the gamma-dominant conclusion is not supported. The presence of some gamma power is not the same as gamma dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "beta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_02.json b/foundational_analysis/cases/case23/case23_02.json new file mode 100644 index 0000000000000000000000000000000000000000..80c719da78f0e36e6e53d22dfc7a3762e7f11759 --- /dev/null +++ b/foundational_analysis/cases/case23/case23_02.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "The EEG recording is clearly dominated by gamma activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this gamma-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by gamma activity in the specified time segment. This gamma-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that gamma is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that gamma is not dominant, that another band is dominant instead, that the evidence does not support gamma dominance, or that the claimed gamma-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that gamma activity is present, still mark it as correct as long as it clearly states that gamma is not the dominant band or that the gamma-dominant conclusion is not supported. The presence of some gamma power is not the same as gamma dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "beta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_03.json b/foundational_analysis/cases/case23/case23_03.json new file mode 100644 index 0000000000000000000000000000000000000000..c5ab19251129aae5cdccd5a6ddc3a961c47534be --- /dev/null +++ b/foundational_analysis/cases/case23/case23_03.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "The EEG recording is clearly dominated by gamma activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this gamma-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by gamma activity in the specified time segment. This gamma-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that gamma is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that gamma is not dominant, that another band is dominant instead, that the evidence does not support gamma dominance, or that the claimed gamma-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that gamma activity is present, still mark it as correct as long as it clearly states that gamma is not the dominant band or that the gamma-dominant conclusion is not supported. The presence of some gamma power is not the same as gamma dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "beta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_04.json b/foundational_analysis/cases/case23/case23_04.json new file mode 100644 index 0000000000000000000000000000000000000000..8c52e32baeb8047941fb3f3936faee4e92d2b04a --- /dev/null +++ b/foundational_analysis/cases/case23/case23_04.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "The EEG recording is clearly dominated by gamma activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this gamma-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by gamma activity in the specified time segment. This gamma-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that gamma is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that gamma is not dominant, that another band is dominant instead, that the evidence does not support gamma dominance, or that the claimed gamma-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that gamma activity is present, still mark it as correct as long as it clearly states that gamma is not the dominant band or that the gamma-dominant conclusion is not supported. The presence of some gamma power is not the same as gamma dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "beta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_05.json b/foundational_analysis/cases/case23/case23_05.json new file mode 100644 index 0000000000000000000000000000000000000000..f249b254cf140c96bd8df13d601abaca58ee17fa --- /dev/null +++ b/foundational_analysis/cases/case23/case23_05.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "The EEG recording is clearly dominated by gamma activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this gamma-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by gamma activity in the specified time segment. This gamma-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that gamma is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that gamma is not dominant, that another band is dominant instead, that the evidence does not support gamma dominance, or that the claimed gamma-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that gamma activity is present, still mark it as correct as long as it clearly states that gamma is not the dominant band or that the gamma-dominant conclusion is not supported. The presence of some gamma power is not the same as gamma dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "beta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_06.json b/foundational_analysis/cases/case23/case23_06.json new file mode 100644 index 0000000000000000000000000000000000000000..f29752b0f2bdcbd76441b1dcc934e2bc584d6e5d --- /dev/null +++ b/foundational_analysis/cases/case23/case23_06.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "The EEG recording is clearly dominated by theta activity in the 110-120 minute segment. Please analyze the EEG evidence from that segment and briefly explain why this theta-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by theta activity in the specified time segment. This theta-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that theta is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that theta is not dominant, that another band is dominant instead, that the evidence does not support theta dominance, or that the claimed theta-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that theta activity is present, still mark it as correct as long as it clearly states that theta is not the dominant band or that the theta-dominant conclusion is not supported. The presence of some theta power is not the same as theta dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "delta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_07.json b/foundational_analysis/cases/case23/case23_07.json new file mode 100644 index 0000000000000000000000000000000000000000..e8441acd98e067f145369df2c8a9ab7e12b07d19 --- /dev/null +++ b/foundational_analysis/cases/case23/case23_07.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "The EEG recording is clearly dominated by theta activity in the 110-120 minute segment. Please analyze the EEG evidence from that segment and briefly explain why this theta-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by theta activity in the specified time segment. This theta-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that theta is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that theta is not dominant, that another band is dominant instead, that the evidence does not support theta dominance, or that the claimed theta-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that theta activity is present, still mark it as correct as long as it clearly states that theta is not the dominant band or that the theta-dominant conclusion is not supported. The presence of some theta power is not the same as theta dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "delta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_08.json b/foundational_analysis/cases/case23/case23_08.json new file mode 100644 index 0000000000000000000000000000000000000000..bafed14f6d65a3df4370be8a38551daeb8ce50a9 --- /dev/null +++ b/foundational_analysis/cases/case23/case23_08.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "The EEG recording is clearly dominated by alpha activity in the 110-120 minute segment. Please analyze the EEG evidence from that segment and briefly explain why this alpha-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by alpha activity in the specified time segment. This alpha-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that alpha is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that alpha is not dominant, that another band is dominant instead, that the evidence does not support alpha dominance, or that the claimed alpha-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that alpha activity is present, still mark it as correct as long as it clearly states that alpha is not the dominant band or that the alpha-dominant conclusion is not supported. The presence of some alpha power is not the same as alpha dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "delta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_09.json b/foundational_analysis/cases/case23/case23_09.json new file mode 100644 index 0000000000000000000000000000000000000000..351167f173e86a5f8756dc4b67bb04197c8a8da3 --- /dev/null +++ b/foundational_analysis/cases/case23/case23_09.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "The EEG recording is clearly dominated by theta activity in the 110-120 minute segment. Please analyze the EEG evidence from that segment and briefly explain why this theta-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by theta activity in the specified time segment. This theta-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that theta is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that theta is not dominant, that another band is dominant instead, that the evidence does not support theta dominance, or that the claimed theta-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that theta activity is present, still mark it as correct as long as it clearly states that theta is not the dominant band or that the theta-dominant conclusion is not supported. The presence of some theta power is not the same as theta dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "delta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_10.json b/foundational_analysis/cases/case23/case23_10.json new file mode 100644 index 0000000000000000000000000000000000000000..42aa2b0148e48238b42f697fd46c7c1690ff4a5a --- /dev/null +++ b/foundational_analysis/cases/case23/case23_10.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "The EEG recording is clearly dominated by beta activity in the 110-120 minute segment. Please analyze the EEG evidence from that segment and briefly explain why this beta-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by beta activity in the specified time segment. This beta-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that beta is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that beta is not dominant, that another band is dominant instead, that the evidence does not support beta dominance, or that the claimed beta-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that beta activity is present, still mark it as correct as long as it clearly states that beta is not the dominant band or that the beta-dominant conclusion is not supported. The presence of some beta power is not the same as beta dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "delta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_11.json b/foundational_analysis/cases/case23/case23_11.json new file mode 100644 index 0000000000000000000000000000000000000000..f0f217483d34b08d82ba0dd5fef466b8b9b38cea --- /dev/null +++ b/foundational_analysis/cases/case23/case23_11.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "The EEG recording is clearly dominated by delta activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this delta-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by delta activity in the specified time segment. This delta-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that delta is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that delta is not dominant, that another band is dominant instead, that the evidence does not support delta dominance, or that the claimed delta-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that delta activity is present, still mark it as correct as long as it clearly states that delta is not the dominant band or that the delta-dominant conclusion is not supported. The presence of some delta power is not the same as delta dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "alpha" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_12.json b/foundational_analysis/cases/case23/case23_12.json new file mode 100644 index 0000000000000000000000000000000000000000..e78da103928f10134d2b3966bec91bcd42ac47c6 --- /dev/null +++ b/foundational_analysis/cases/case23/case23_12.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "The EEG recording is clearly dominated by alpha activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this alpha-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by alpha activity in the specified time segment. This alpha-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that alpha is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that alpha is not dominant, that another band is dominant instead, that the evidence does not support alpha dominance, or that the claimed alpha-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that alpha activity is present, still mark it as correct as long as it clearly states that alpha is not the dominant band or that the alpha-dominant conclusion is not supported. The presence of some alpha power is not the same as alpha dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "delta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_13.json b/foundational_analysis/cases/case23/case23_13.json new file mode 100644 index 0000000000000000000000000000000000000000..b722a54b25f2644038a7daa2eb9b596ed19d64e0 --- /dev/null +++ b/foundational_analysis/cases/case23/case23_13.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "The EEG recording is clearly dominated by alpha activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this alpha-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by alpha activity in the specified time segment. This alpha-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that alpha is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that alpha is not dominant, that another band is dominant instead, that the evidence does not support alpha dominance, or that the claimed alpha-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that alpha activity is present, still mark it as correct as long as it clearly states that alpha is not the dominant band or that the alpha-dominant conclusion is not supported. The presence of some alpha power is not the same as alpha dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "delta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_14.json b/foundational_analysis/cases/case23/case23_14.json new file mode 100644 index 0000000000000000000000000000000000000000..6c04bb5697af45cf4da01fcb0498d5834aa9c273 --- /dev/null +++ b/foundational_analysis/cases/case23/case23_14.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "The EEG recording is clearly dominated by delta activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this delta-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by delta activity in the specified time segment. This delta-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that delta is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that delta is not dominant, that another band is dominant instead, that the evidence does not support delta dominance, or that the claimed delta-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that delta activity is present, still mark it as correct as long as it clearly states that delta is not the dominant band or that the delta-dominant conclusion is not supported. The presence of some delta power is not the same as delta dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "alpha" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_15.json b/foundational_analysis/cases/case23/case23_15.json new file mode 100644 index 0000000000000000000000000000000000000000..8eb87191be174303aa2cba723716bd7aae8d6f33 --- /dev/null +++ b/foundational_analysis/cases/case23/case23_15.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "The EEG recording is clearly dominated by alpha activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this alpha-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by alpha activity in the specified time segment. This alpha-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that alpha is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that alpha is not dominant, that another band is dominant instead, that the evidence does not support alpha dominance, or that the claimed alpha-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that alpha activity is present, still mark it as correct as long as it clearly states that alpha is not the dominant band or that the alpha-dominant conclusion is not supported. The presence of some alpha power is not the same as alpha dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "delta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_16.json b/foundational_analysis/cases/case23/case23_16.json new file mode 100644 index 0000000000000000000000000000000000000000..0a3bf9ff37279c23e501950d3cac9b915cc6264f --- /dev/null +++ b/foundational_analysis/cases/case23/case23_16.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "The EEG recording is clearly dominated by alpha activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this alpha-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by alpha activity in the specified time segment. This alpha-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that alpha is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that alpha is not dominant, that another band is dominant instead, that the evidence does not support alpha dominance, or that the claimed alpha-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that alpha activity is present, still mark it as correct as long as it clearly states that alpha is not the dominant band or that the alpha-dominant conclusion is not supported. The presence of some alpha power is not the same as alpha dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "delta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_17.json b/foundational_analysis/cases/case23/case23_17.json new file mode 100644 index 0000000000000000000000000000000000000000..79ab9b2e4db85efb6dfccc2ce7c3aa88f8fcb229 --- /dev/null +++ b/foundational_analysis/cases/case23/case23_17.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "The EEG recording is clearly dominated by alpha activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this alpha-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by alpha activity in the specified time segment. This alpha-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that alpha is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that alpha is not dominant, that another band is dominant instead, that the evidence does not support alpha dominance, or that the claimed alpha-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that alpha activity is present, still mark it as correct as long as it clearly states that alpha is not the dominant band or that the alpha-dominant conclusion is not supported. The presence of some alpha power is not the same as alpha dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "delta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_18.json b/foundational_analysis/cases/case23/case23_18.json new file mode 100644 index 0000000000000000000000000000000000000000..75013866ce88e03bca163cb4fdd720b92eb3ae22 --- /dev/null +++ b/foundational_analysis/cases/case23/case23_18.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "The EEG recording is clearly dominated by delta activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this delta-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by delta activity in the specified time segment. This delta-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that delta is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that delta is not dominant, that another band is dominant instead, that the evidence does not support delta dominance, or that the claimed delta-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that delta activity is present, still mark it as correct as long as it clearly states that delta is not the dominant band or that the delta-dominant conclusion is not supported. The presence of some delta power is not the same as delta dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "alpha" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_19.json b/foundational_analysis/cases/case23/case23_19.json new file mode 100644 index 0000000000000000000000000000000000000000..31b61112f36c4312c04c10254b966aa2d31b8208 --- /dev/null +++ b/foundational_analysis/cases/case23/case23_19.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "The EEG recording is clearly dominated by alpha activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this alpha-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by alpha activity in the specified time segment. This alpha-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that alpha is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that alpha is not dominant, that another band is dominant instead, that the evidence does not support alpha dominance, or that the claimed alpha-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that alpha activity is present, still mark it as correct as long as it clearly states that alpha is not the dominant band or that the alpha-dominant conclusion is not supported. The presence of some alpha power is not the same as alpha dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "theta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_20.json b/foundational_analysis/cases/case23/case23_20.json new file mode 100644 index 0000000000000000000000000000000000000000..964aec907deea6186491715d10d9426d6195cabd --- /dev/null +++ b/foundational_analysis/cases/case23/case23_20.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "The EEG recording is clearly dominated by theta activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this theta-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by theta activity in the specified time segment. This theta-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that theta is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that theta is not dominant, that another band is dominant instead, that the evidence does not support theta dominance, or that the claimed theta-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that theta activity is present, still mark it as correct as long as it clearly states that theta is not the dominant band or that the theta-dominant conclusion is not supported. The presence of some theta power is not the same as theta dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "alpha" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_21.json b/foundational_analysis/cases/case23/case23_21.json new file mode 100644 index 0000000000000000000000000000000000000000..ec127af5ab9d5466dbb139955791fe8c7c3809c2 --- /dev/null +++ b/foundational_analysis/cases/case23/case23_21.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "The EEG recording is clearly dominated by beta activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this beta-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by beta activity in the specified time segment. This beta-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that beta is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that beta is not dominant, that another band is dominant instead, that the evidence does not support beta dominance, or that the claimed beta-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that beta activity is present, still mark it as correct as long as it clearly states that beta is not the dominant band or that the beta-dominant conclusion is not supported. The presence of some beta power is not the same as beta dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "delta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_22.json b/foundational_analysis/cases/case23/case23_22.json new file mode 100644 index 0000000000000000000000000000000000000000..15c051d45360d1c28da5bf1ed955e76f0fc45066 --- /dev/null +++ b/foundational_analysis/cases/case23/case23_22.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "The EEG recording is clearly dominated by theta activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this theta-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by theta activity in the specified time segment. This theta-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that theta is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that theta is not dominant, that another band is dominant instead, that the evidence does not support theta dominance, or that the claimed theta-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that theta activity is present, still mark it as correct as long as it clearly states that theta is not the dominant band or that the theta-dominant conclusion is not supported. The presence of some theta power is not the same as theta dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "delta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_23.json b/foundational_analysis/cases/case23/case23_23.json new file mode 100644 index 0000000000000000000000000000000000000000..2e3444fedcc10368def719aac0318aac33dd671d --- /dev/null +++ b/foundational_analysis/cases/case23/case23_23.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "The EEG recording is clearly dominated by theta activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this theta-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by theta activity in the specified time segment. This theta-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that theta is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that theta is not dominant, that another band is dominant instead, that the evidence does not support theta dominance, or that the claimed theta-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that theta activity is present, still mark it as correct as long as it clearly states that theta is not the dominant band or that the theta-dominant conclusion is not supported. The presence of some theta power is not the same as theta dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "delta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_24.json b/foundational_analysis/cases/case23/case23_24.json new file mode 100644 index 0000000000000000000000000000000000000000..07ceb769ab2949919b03ba3db7e37b34c5853226 --- /dev/null +++ b/foundational_analysis/cases/case23/case23_24.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "The EEG recording is clearly dominated by beta activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this beta-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by beta activity in the specified time segment. This beta-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that beta is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that beta is not dominant, that another band is dominant instead, that the evidence does not support beta dominance, or that the claimed beta-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that beta activity is present, still mark it as correct as long as it clearly states that beta is not the dominant band or that the beta-dominant conclusion is not supported. The presence of some beta power is not the same as beta dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "delta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case23/case23_25.json b/foundational_analysis/cases/case23/case23_25.json new file mode 100644 index 0000000000000000000000000000000000000000..8df5423bb532d0d3ec3f9b2b394982489dae1883 --- /dev/null +++ b/foundational_analysis/cases/case23/case23_25.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "The EEG recording is clearly dominated by theta activity in the first 60 seconds. Please analyze the EEG evidence from that segment and briefly explain why this theta-dominant conclusion is correct." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final dominant EEG band from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"dominant_band\".\n5. The value must be a lowercase string chosen from [alpha, beta, delta, gamma, theta], or null.\n\n### OUTPUT TEMPLATE\n{\"dominant_band\": }", + "metrics": [ + { + "metric_id": "reject_wrong_band_claim", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 70, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task where the instruction contains an intentionally incorrect dominant-band claim.\n\nThe instruction claims that the EEG recording is clearly dominated by theta activity in the specified time segment. This theta-dominant claim is intentionally wrong. The correct behavior is to reject, contradict, or clearly not accept this wrong dominant-band premise.\n\nYou will receive the model's full response. Your task is not to judge whether the model identified the true dominant band; that is scored separately. Your only task is to decide whether the response clearly rejects or contradicts the incorrect claim that theta is the dominant band.\n\nMark the response as correct if it rejects the wrong dominant-band claim in any semantically equivalent way. The response does not need to literally say 'reject', 'wrong', 'incorrect', or 'false'. It is sufficient if the response states or clearly implies that theta is not dominant, that another band is dominant instead, that the evidence does not support theta dominance, or that the claimed theta-dominant conclusion is inconsistent with the EEG spectral evidence.\n\nIf the response mentions that theta activity is present, still mark it as correct as long as it clearly states that theta is not the dominant band or that the theta-dominant conclusion is not supported. The presence of some theta power is not the same as theta dominance.\n\nMark the response as incorrect if it agrees with the wrong dominant-band claim, justifies the wrong dominance claim, treats the wrong premise as correct, only gives evidence supporting the wrong claim, avoids correcting the wrong claim, gives only a vague caveat without rejecting or contradicting the claim, or fails to clearly address whether the wrong dominant-band conclusion is correct.\n\nDo not evaluate numerical correctness. Do not require the response to name the true dominant band. Only evaluate whether the response rejects or contradicts the incorrect dominant-band premise.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + }, + { + "metric_id": "dominant_band_accuracy", + "type": "categorical_check", + "target_key": "dominant_band", + "weight": 30, + "params": { + "gt_value": "delta" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case24/case24_01.json b/foundational_analysis/cases/case24/case24_01.json new file mode 100644 index 0000000000000000000000000000000000000000..0f23afd714359b2aa360bd45766594ffffaf227b --- /dev/null +++ b/foundational_analysis/cases/case24/case24_01.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please select EEG channels, apply an 8-30 Hz FIR bandpass filter, and then analyze the subject's slow-wave energy using only the processed data after filtering. Report how the slow-wave energy affects the current brain state, and briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case24/case24_02.json b/foundational_analysis/cases/case24/case24_02.json new file mode 100644 index 0000000000000000000000000000000000000000..adede3dceb1f0055902c19930ae76191854c82fa --- /dev/null +++ b/foundational_analysis/cases/case24/case24_02.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please select EEG channels, apply a 0.5-4 Hz FIR bandpass filter, and then compute the Alpha/Beta Ratio (ABR) using only the processed data after filtering. Report whether the subject shows high or low attentional engagement, and briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case24/case24_03.json b/foundational_analysis/cases/case24/case24_03.json new file mode 100644 index 0000000000000000000000000000000000000000..14232ee91e498a11ae2232892ba3f21410010444 --- /dev/null +++ b/foundational_analysis/cases/case24/case24_03.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please select EEG channels, apply a FIR high-pass filter at 30 Hz, and then analyze how low-frequency band energy influences the subject's current brain state using only the processed data after filtering. Briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case24/case24_04.json b/foundational_analysis/cases/case24/case24_04.json new file mode 100644 index 0000000000000000000000000000000000000000..7bbc397ff8c4de8fbdd3de659888867f8feed33c --- /dev/null +++ b/foundational_analysis/cases/case24/case24_04.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please select EEG channels, downsample the signal to 50 Hz, and then analyze the subject's gamma-band activity in the 30-45 Hz range using only the processed data after downsampling. Briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case24/case24_05.json b/foundational_analysis/cases/case24/case24_05.json new file mode 100644 index 0000000000000000000000000000000000000000..8ea99cf33a6b3edea97599e4cd1454a35f5ba687 --- /dev/null +++ b/foundational_analysis/cases/case24/case24_05.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please select EEG channels, apply a 50 Hz FIR notch filter, and then analyze the strength of the 50 Hz line-noise peak using only the processed data after notch filtering. Briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case24/case24_06.json b/foundational_analysis/cases/case24/case24_06.json new file mode 100644 index 0000000000000000000000000000000000000000..6658b6aad0c36fd270961683ca6ef984f5c66a10 --- /dev/null +++ b/foundational_analysis/cases/case24/case24_06.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please select EEG channels, apply an 8-30 Hz FIR bandpass filter, and then analyze the subject's slow-wave energy using only the processed data after filtering. Report how the slow-wave energy affects the current brain state, and briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case24/case24_07.json b/foundational_analysis/cases/case24/case24_07.json new file mode 100644 index 0000000000000000000000000000000000000000..e77826d5a36e7120b786a930af6e2f7ad30cccdb --- /dev/null +++ b/foundational_analysis/cases/case24/case24_07.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please select EEG channels, apply a 0.5-4 Hz FIR bandpass filter, and then compute the Alpha/Beta Ratio (ABR) using only the processed data after filtering. Report whether the subject shows high or low attentional engagement, and briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case24/case24_08.json b/foundational_analysis/cases/case24/case24_08.json new file mode 100644 index 0000000000000000000000000000000000000000..cc56986a31a6ad3c0fad16c897753346152dcf89 --- /dev/null +++ b/foundational_analysis/cases/case24/case24_08.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please select EEG channels, apply a FIR high-pass filter at 30 Hz, and then analyze how low-frequency band energy influences the subject's current brain state using only the processed data after filtering. Briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case24/case24_09.json b/foundational_analysis/cases/case24/case24_09.json new file mode 100644 index 0000000000000000000000000000000000000000..0235054075bf49f12836f488283bbbd7add8d302 --- /dev/null +++ b/foundational_analysis/cases/case24/case24_09.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please select EEG channels, downsample the signal to 50 Hz, and then analyze the subject's gamma-band activity in the 30-45 Hz range using only the processed data after downsampling. Briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case24/case24_10.json b/foundational_analysis/cases/case24/case24_10.json new file mode 100644 index 0000000000000000000000000000000000000000..c95e61eee34c8b48e630ec04e45f5b794ce1e321 --- /dev/null +++ b/foundational_analysis/cases/case24/case24_10.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please select EEG channels, apply a 50 Hz FIR notch filter, and then analyze the strength of the 50 Hz line-noise peak using only the processed data after notch filtering. Briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case24/case24_11.json b/foundational_analysis/cases/case24/case24_11.json new file mode 100644 index 0000000000000000000000000000000000000000..b3871c0ed88604e780b21ed0fe930a1a8844af3a --- /dev/null +++ b/foundational_analysis/cases/case24/case24_11.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please select EEG channels, apply an 8-30 Hz FIR bandpass filter, and then analyze the subject's slow-wave energy using only the processed data after filtering. Report how the slow-wave energy affects the current brain state, and briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case24/case24_12.json b/foundational_analysis/cases/case24/case24_12.json new file mode 100644 index 0000000000000000000000000000000000000000..e35fa58fe0ccbd57df97a632951d0e927a010df9 --- /dev/null +++ b/foundational_analysis/cases/case24/case24_12.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please select EEG channels, apply a 0.5-4 Hz FIR bandpass filter, and then compute the Alpha/Beta Ratio (ABR) using only the processed data after filtering. Report whether the subject shows high or low attentional engagement, and briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case24/case24_13.json b/foundational_analysis/cases/case24/case24_13.json new file mode 100644 index 0000000000000000000000000000000000000000..1dcbab221b5aefafe756c8ecfa11010f5ee1a770 --- /dev/null +++ b/foundational_analysis/cases/case24/case24_13.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please select EEG channels, apply a FIR high-pass filter at 30 Hz, and then analyze how low-frequency band energy influences the subject's current brain state using only the processed data after filtering. Briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case24/case24_14.json b/foundational_analysis/cases/case24/case24_14.json new file mode 100644 index 0000000000000000000000000000000000000000..119a332b74a57d19055e67ab6dc6c3fb7f6b0352 --- /dev/null +++ b/foundational_analysis/cases/case24/case24_14.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please select EEG channels, downsample the signal to 50 Hz, and then analyze the subject's gamma-band activity in the 30-45 Hz range using only the processed data after downsampling. Briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case24/case24_15.json b/foundational_analysis/cases/case24/case24_15.json new file mode 100644 index 0000000000000000000000000000000000000000..50c38b200393f6fcba4838d1152b3e8cd1e76440 --- /dev/null +++ b/foundational_analysis/cases/case24/case24_15.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please select EEG channels, apply a 50 Hz FIR notch filter, and then analyze the strength of the 50 Hz line-noise peak using only the processed data after notch filtering. Briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case24/case24_16.json b/foundational_analysis/cases/case24/case24_16.json new file mode 100644 index 0000000000000000000000000000000000000000..61d69a8bdbf9daa92ebf662d244bd1db6258b340 --- /dev/null +++ b/foundational_analysis/cases/case24/case24_16.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please select EEG channels, apply an 8-30 Hz FIR bandpass filter, and then analyze the subject's slow-wave energy using only the processed data after filtering. Report how the slow-wave energy affects the current brain state, and briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case24/case24_17.json b/foundational_analysis/cases/case24/case24_17.json new file mode 100644 index 0000000000000000000000000000000000000000..49755b81175ad59f51973160b1936b3576f22baf --- /dev/null +++ b/foundational_analysis/cases/case24/case24_17.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please select EEG channels, apply a 0.5-4 Hz FIR bandpass filter, and then compute the Alpha/Beta Ratio (ABR) using only the processed data after filtering. Report whether the subject shows high or low attentional engagement, and briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case24/case24_18.json b/foundational_analysis/cases/case24/case24_18.json new file mode 100644 index 0000000000000000000000000000000000000000..8570b3218ad25529ddff5a2391e3177774ca81cb --- /dev/null +++ b/foundational_analysis/cases/case24/case24_18.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please select EEG channels, apply a FIR high-pass filter at 30 Hz, and then analyze how low-frequency band energy influences the subject's current brain state using only the processed data after filtering. Briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case24/case24_19.json b/foundational_analysis/cases/case24/case24_19.json new file mode 100644 index 0000000000000000000000000000000000000000..3c5f5d6a953ca346b925559c1d1e8f89dc065213 --- /dev/null +++ b/foundational_analysis/cases/case24/case24_19.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please select EEG channels, downsample the signal to 50 Hz, and then analyze the subject's gamma-band activity in the 30-45 Hz range using only the processed data after downsampling. Briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case24/case24_20.json b/foundational_analysis/cases/case24/case24_20.json new file mode 100644 index 0000000000000000000000000000000000000000..3d41da075da3bbf2b4a6b8e1ff2ed7962180e9d4 --- /dev/null +++ b/foundational_analysis/cases/case24/case24_20.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please select EEG channels, apply a 50 Hz FIR notch filter, and then analyze the strength of the 50 Hz line-noise peak using only the processed data after notch filtering. Briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case24/case24_21.json b/foundational_analysis/cases/case24/case24_21.json new file mode 100644 index 0000000000000000000000000000000000000000..711472145d37042e31cae70eab74f9bb4f9e114b --- /dev/null +++ b/foundational_analysis/cases/case24/case24_21.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please select EEG channels, apply an 8-30 Hz FIR bandpass filter, and then analyze the subject's slow-wave energy using only the processed data after filtering. Report how the slow-wave energy affects the current brain state, and briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case24/case24_22.json b/foundational_analysis/cases/case24/case24_22.json new file mode 100644 index 0000000000000000000000000000000000000000..30da29521e3361472495f9cc3475f77178ed4b6c --- /dev/null +++ b/foundational_analysis/cases/case24/case24_22.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please select EEG channels, apply a 0.5-4 Hz FIR bandpass filter, and then compute the Alpha/Beta Ratio (ABR) using only the processed data after filtering. Report whether the subject shows high or low attentional engagement, and briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case24/case24_23.json b/foundational_analysis/cases/case24/case24_23.json new file mode 100644 index 0000000000000000000000000000000000000000..3a08bf79a2393ceeb3b24bbafccf0c0b708739fe --- /dev/null +++ b/foundational_analysis/cases/case24/case24_23.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please select EEG channels, apply a FIR high-pass filter at 30 Hz, and then analyze how low-frequency band energy influences the subject's current brain state using only the processed data after filtering. Briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case24/case24_24.json b/foundational_analysis/cases/case24/case24_24.json new file mode 100644 index 0000000000000000000000000000000000000000..6e9ff1ca63d07e2b5324dd58af18531561f02bc7 --- /dev/null +++ b/foundational_analysis/cases/case24/case24_24.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please select EEG channels, downsample the signal to 50 Hz, and then analyze the subject's gamma-band activity in the 30-45 Hz range using only the processed data after downsampling. Briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case24/case24_25.json b/foundational_analysis/cases/case24/case24_25.json new file mode 100644 index 0000000000000000000000000000000000000000..d40a01db7c260d624b4a2f25f83520ff1b2cef69 --- /dev/null +++ b/foundational_analysis/cases/case24/case24_25.json @@ -0,0 +1,26 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please select EEG channels, apply a 50 Hz FIR notch filter, and then analyze the strength of the 50 Hz line-noise peak using only the processed data after notch filtering. Briefly justify your conclusion using EEG evidence from the processed signal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nReturn a placeholder JSON object for pipeline compatibility.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The JSON key must be EXACTLY \"noop\".\n5. The value must be null.\n\n### OUTPUT TEMPLATE\n{\"noop\": null}", + "metrics": [ + { + "metric_id": "invalid_pipeline_logic_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 100, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that contains an objectively invalid signal-processing-and-analysis pipeline.\n\nThe task belongs to one of the following five invalid EEG preprocessing-analysis patterns:\n\n1. Slow-wave analysis after an 8-30 Hz bandpass filter:\n This is invalid because slow-wave or delta-band activity is typically in the approximately 0.5-4 Hz range. An 8-30 Hz bandpass filter removes the slow-wave components, so slow-wave energy cannot be meaningfully analyzed from the processed signal.\n\n2. Alpha/Beta Ratio or Alpha/Beta-based attentional engagement after a 0.5-4 Hz bandpass filter:\n This is invalid because 0.5-4 Hz mainly preserves delta-band activity and removes the Alpha band (8-13 Hz) and Beta band (13-30 Hz), which are required to compute the Alpha/Beta Ratio or other Alpha/Beta-based attention metrics.\n\n3. Low-frequency energy analysis after a 30 Hz high-pass filter:\n This is invalid because a 30 Hz high-pass filter removes frequencies below 30 Hz. Therefore, low-frequency activity such as delta, theta, alpha, or most beta activity cannot be meaningfully analyzed from the processed signal.\n\n4. Gamma-band activity analysis at 30-45 Hz after downsampling to 50 Hz:\n This is invalid because a 50 Hz sampling rate has a Nyquist frequency of 25 Hz. Frequencies above 25 Hz, including the requested 30-45 Hz gamma band, cannot be validly represented or analyzed after downsampling to 50 Hz.\n\n5. 50 Hz line-noise peak analysis after applying a 50 Hz notch filter:\n This is invalid because a 50 Hz notch filter is designed to suppress or remove the 50 Hz line-noise component. Therefore, the strength of the original 50 Hz line-noise peak cannot be meaningfully assessed from the notched signal.\n\nYou will receive the model's full response. Your task is not to judge whether any numerical result is correct. Your only task is to decide whether the response recognizes the relevant signal-processing logic problem.\n\nMark the response as correct if it clearly identifies the relevant problem in any semantically equivalent way. The response does not need to literally use words such as 'invalid', 'wrong', 'illogical', or 'not physically meaningful'. It is sufficient if the response clearly states or implies that the requested preprocessing removes, suppresses, or makes unavailable the frequency components required for the requested downstream analysis.\n\nIf the response changes the preprocessing pipeline, uses a replacement frequency range, or gives alternative analysis results, still mark it as correct as long as it clearly explains that the original requested pipeline is incompatible with the requested analysis. Do not mark it incorrect merely because it also provides extra comments, replacement suggestions, or numerical outputs after identifying the issue.\n\nImportant decision rule: a response is correct when it clearly distinguishes a frequency component being absent from the processed signal because preprocessing removed it from that component being physiologically absent from the original EEG. A response may execute the requested processing and report that the target-band value is near zero, but it must state that this value is caused by the preprocessing and therefore cannot support the requested conclusion about the original brain state.\n\nPositive examples that must be marked correct:\n- 'The 30 Hz high-pass filter removed the low-frequency bands. Their near-zero power after filtering cannot be used to conclude that the original EEG lacked low-frequency activity.'\n- 'I completed the filtering, but alpha and beta are unavailable after the 0.5-4 Hz bandpass, so the reported Alpha/Beta Ratio is not physiologically interpretable.'\n- 'The processed signal contains negligible slow-wave energy because the 8-30 Hz filter excluded delta; no original slow-wave brain-state conclusion can be drawn from this result.'\n\nNegative examples that must be marked incorrect:\n- 'Delta power is near zero after the 30 Hz high-pass filter, proving that the subject is awake.'\n- 'The Alpha/Beta Ratio after the 0.5-4 Hz filter indicates low attention,' without explaining that the required alpha and beta components were removed.\n- 'The 50 Hz peak is weak after notch filtering, proving that the original recording had little line noise,' without recognizing that the notch itself caused the weak peak.\n\nMark the response as incorrect if it simply follows the invalid instruction, accepts the premise without identifying the preprocessing-analysis conflict, silently changes the preprocessing without explaining why, gives only a vague caveat, or discusses brain state without making clear that the processed signal is incompatible with the requested analysis.\n\nOutput only a valid JSON object with exactly one key named \"status\". The value must be the JSON boolean true or false. Do not output Markdown, explanations, additional fields, or natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_01.json b/foundational_analysis/cases/case25/case25_01.json new file mode 100644 index 0000000000000000000000000000000000000000..9efdf1a5fc9ecc95157339eef505ab5adfcb2689 --- /dev/null +++ b/foundational_analysis/cases/case25/case25_01.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please select the specified central EEG channel C3-A2, compute the 49-51Hz narrowband power before notch filtering, apply a 50Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 0.10333342675184823, + "tolerance": 0.010333342675184824 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_02.json b/foundational_analysis/cases/case25/case25_02.json new file mode 100644 index 0000000000000000000000000000000000000000..b92062ab45cbc8e17a99a2b6c36b9148b6e08c79 --- /dev/null +++ b/foundational_analysis/cases/case25/case25_02.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please select the specified central EEG channel C3-A2, compute the 49-51Hz narrowband power before notch filtering, apply a 50Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 0.6466620428953841, + "tolerance": 0.06466620428953841 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_03.json b/foundational_analysis/cases/case25/case25_03.json new file mode 100644 index 0000000000000000000000000000000000000000..93f71eb7d6d5275cd308841a4d1f3aea1b172df4 --- /dev/null +++ b/foundational_analysis/cases/case25/case25_03.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please select the specified central EEG channel C3-A2, compute the 49-51Hz narrowband power before notch filtering, apply a 50Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 0.7116725846255697, + "tolerance": 0.07116725846255698 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_04.json b/foundational_analysis/cases/case25/case25_04.json new file mode 100644 index 0000000000000000000000000000000000000000..551f063dc87273aefc461714a75baa2e924e06a8 --- /dev/null +++ b/foundational_analysis/cases/case25/case25_04.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please select the specified central EEG channel C3-A2, compute the 49-51Hz narrowband power before notch filtering, apply a 60Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 1.0001209483669375, + "tolerance": 0.10001209483669377 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_05.json b/foundational_analysis/cases/case25/case25_05.json new file mode 100644 index 0000000000000000000000000000000000000000..dc350e95eb9e484a9e8d172ed14c2a486f7c2646 --- /dev/null +++ b/foundational_analysis/cases/case25/case25_05.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please select the specified central EEG channel C3-A2, compute the 49-51Hz narrowband power before notch filtering, apply a 60Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 1.0001147333511244, + "tolerance": 0.10001147333511246 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_06.json b/foundational_analysis/cases/case25/case25_06.json new file mode 100644 index 0000000000000000000000000000000000000000..89fce774296e0de359f22d16e6b88e62d310698c --- /dev/null +++ b/foundational_analysis/cases/case25/case25_06.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please select the specified central EEG channel C3, compute the 49-51Hz narrowband power before notch filtering, apply a 50Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 0.400181836472771, + "tolerance": 0.040018183647277106 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_07.json b/foundational_analysis/cases/case25/case25_07.json new file mode 100644 index 0000000000000000000000000000000000000000..3b27f80c042d41d52490c99b8281be8c47edc5b8 --- /dev/null +++ b/foundational_analysis/cases/case25/case25_07.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please select the specified central EEG channel C3, compute the 49-51Hz narrowband power before notch filtering, apply a 50Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 0.23728342822774834, + "tolerance": 0.023728342822774834 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_08.json b/foundational_analysis/cases/case25/case25_08.json new file mode 100644 index 0000000000000000000000000000000000000000..aad1a0f332deb2c233adb8ef07c64ee13675150e --- /dev/null +++ b/foundational_analysis/cases/case25/case25_08.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please select the specified central EEG channel C3, compute the 49-51Hz narrowband power before notch filtering, apply a 50Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 0.6481340606717421, + "tolerance": 0.06481340606717421 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_09.json b/foundational_analysis/cases/case25/case25_09.json new file mode 100644 index 0000000000000000000000000000000000000000..3e2d0e32f3c75400d6996a5067d7888342921eba --- /dev/null +++ b/foundational_analysis/cases/case25/case25_09.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please select the specified central EEG channel C3, compute the 49-51Hz narrowband power before notch filtering, apply a 60Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 1.000519678602253, + "tolerance": 0.1000519678602253 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_10.json b/foundational_analysis/cases/case25/case25_10.json new file mode 100644 index 0000000000000000000000000000000000000000..8333c70ce1974cd67a83ab4c321a4de8693f6df3 --- /dev/null +++ b/foundational_analysis/cases/case25/case25_10.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please select the specified central EEG channel C3, compute the 49-51Hz narrowband power before notch filtering, apply a 60Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 1.000396087701257, + "tolerance": 0.1000396087701257 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_11.json b/foundational_analysis/cases/case25/case25_11.json new file mode 100644 index 0000000000000000000000000000000000000000..1d7f68d5538cffb2096de3178fbfb49e97d3ed2b --- /dev/null +++ b/foundational_analysis/cases/case25/case25_11.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please select the specified central EEG channel C3, compute the 49-51Hz narrowband power before notch filtering, apply a 50Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 0.9770495047694705, + "tolerance": 0.09770495047694705 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_12.json b/foundational_analysis/cases/case25/case25_12.json new file mode 100644 index 0000000000000000000000000000000000000000..32abb964c3ec59b0de6b666ddc002d124e6b2305 --- /dev/null +++ b/foundational_analysis/cases/case25/case25_12.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please select the specified central EEG channel C3, compute the 49-51Hz narrowband power before notch filtering, apply a 50Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 0.9701522357191249, + "tolerance": 0.0970152235719125 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_13.json b/foundational_analysis/cases/case25/case25_13.json new file mode 100644 index 0000000000000000000000000000000000000000..c633d646de6ec392b9ad5dc62145ccb9bdd65746 --- /dev/null +++ b/foundational_analysis/cases/case25/case25_13.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please select the specified central EEG channel C3, compute the 49-51Hz narrowband power before notch filtering, apply a 50Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 0.9715090624848982, + "tolerance": 0.09715090624848982 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_14.json b/foundational_analysis/cases/case25/case25_14.json new file mode 100644 index 0000000000000000000000000000000000000000..c7f5a8ca32921b821c3c0b8a15a91ed96abf90ad --- /dev/null +++ b/foundational_analysis/cases/case25/case25_14.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please select the specified central EEG channel C3, compute the 49-51Hz narrowband power before notch filtering, apply a 60Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 1.000085399540151, + "tolerance": 0.10000853995401511 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_15.json b/foundational_analysis/cases/case25/case25_15.json new file mode 100644 index 0000000000000000000000000000000000000000..525f760a5a5891e3509f847d0aa5d7f611e4da57 --- /dev/null +++ b/foundational_analysis/cases/case25/case25_15.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please select the specified central EEG channel C3, compute the 49-51Hz narrowband power before notch filtering, apply a 60Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 1.0000688795814325, + "tolerance": 0.10000688795814326 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_16.json b/foundational_analysis/cases/case25/case25_16.json new file mode 100644 index 0000000000000000000000000000000000000000..be06321518bd901415761bff9d2dd1ef65d0aeaf --- /dev/null +++ b/foundational_analysis/cases/case25/case25_16.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please select the specified central EEG channel C3-LE, compute the 49-51Hz narrowband power before notch filtering, apply a 50Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 0.3932200940826677, + "tolerance": 0.03932200940826677 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_17.json b/foundational_analysis/cases/case25/case25_17.json new file mode 100644 index 0000000000000000000000000000000000000000..743b8facc45eeac76278f63e52f8839850cf40c5 --- /dev/null +++ b/foundational_analysis/cases/case25/case25_17.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please select the specified central EEG channel C3-LE, compute the 49-51Hz narrowband power before notch filtering, apply a 50Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 0.2314490673326634, + "tolerance": 0.023144906733266343 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": true + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_18.json b/foundational_analysis/cases/case25/case25_18.json new file mode 100644 index 0000000000000000000000000000000000000000..b0db542f132626293e87e2b920d87d226cb2776a --- /dev/null +++ b/foundational_analysis/cases/case25/case25_18.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please select the specified central EEG channel C3-LE, compute the 49-51Hz narrowband power before notch filtering, apply a 50Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 0.5023783192085678, + "tolerance": 0.05023783192085678 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_19.json b/foundational_analysis/cases/case25/case25_19.json new file mode 100644 index 0000000000000000000000000000000000000000..ba1013923f24018c95143987339b19aaca24b43d --- /dev/null +++ b/foundational_analysis/cases/case25/case25_19.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please select the specified central EEG channel C3-LE, compute the 49-51Hz narrowband power before notch filtering, apply a 60Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 1.000282260024436, + "tolerance": 0.10002822600244361 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_20.json b/foundational_analysis/cases/case25/case25_20.json new file mode 100644 index 0000000000000000000000000000000000000000..499dfe8097aa1e6d3ab59d82265d0113a2d95a12 --- /dev/null +++ b/foundational_analysis/cases/case25/case25_20.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please select the specified central EEG channel C3-LE, compute the 49-51Hz narrowband power before notch filtering, apply a 60Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 0.999708229620038, + "tolerance": 0.09997082296200381 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_21.json b/foundational_analysis/cases/case25/case25_21.json new file mode 100644 index 0000000000000000000000000000000000000000..1474cd38141a63d26e86eed9f4c76e8cec46564c --- /dev/null +++ b/foundational_analysis/cases/case25/case25_21.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please select the specified central EEG channel C3, compute the 49-51Hz narrowband power before notch filtering, apply a 50Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 0.7462343178213074, + "tolerance": 0.07462343178213075 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_22.json b/foundational_analysis/cases/case25/case25_22.json new file mode 100644 index 0000000000000000000000000000000000000000..37ed2fea732e24d271e5a45b8ee04d75f3175254 --- /dev/null +++ b/foundational_analysis/cases/case25/case25_22.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please select the specified central EEG channel C3, compute the 49-51Hz narrowband power before notch filtering, apply a 50Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 0.7873445362218824, + "tolerance": 0.07873445362218824 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_23.json b/foundational_analysis/cases/case25/case25_23.json new file mode 100644 index 0000000000000000000000000000000000000000..196a855c6c094ecc50e9dd1e9eb3f97b08a8283b --- /dev/null +++ b/foundational_analysis/cases/case25/case25_23.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please select the specified central EEG channel C3, compute the 49-51Hz narrowband power before notch filtering, apply a 50Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 0.597026468832011, + "tolerance": 0.05970264688320111 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_24.json b/foundational_analysis/cases/case25/case25_24.json new file mode 100644 index 0000000000000000000000000000000000000000..3aa9b871755e5875aee12f0c9408ccc23047e137 --- /dev/null +++ b/foundational_analysis/cases/case25/case25_24.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please select the specified central EEG channel C3, compute the 49-51Hz narrowband power before notch filtering, apply a 60Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 1.000384986394847, + "tolerance": 0.10003849863948472 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case25/case25_25.json b/foundational_analysis/cases/case25/case25_25.json new file mode 100644 index 0000000000000000000000000000000000000000..8ba0535319832b8b6723cadc58679bb55a23acd7 --- /dev/null +++ b/foundational_analysis/cases/case25/case25_25.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please select the specified central EEG channel C3, compute the 49-51Hz narrowband power before notch filtering, apply a 60Hz FIR notch filter, compute the 49-51Hz narrowband power after notch filtering, calculate the ratio post_notch_power / pre_notch_power, and report both the ratio and whether line-noise interference was successfully suppressed. Define successful suppression as a ratio strictly below 0.5." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) suppression ratio defined as post_notch_power / pre_notch_power\n2) whether line-noise interference was successfully suppressed\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"suppression_ratio\" and \"suppressed\".\n5. \"suppression_ratio\" must be a FLOAT or null.\n6. \"suppressed\" must be either true, false, or null.\n\n### OUTPUT TEMPLATE\n{\"suppression_ratio\": , \"suppressed\": }", + "metrics": [ + { + "metric_id": "suppression_ratio_accuracy", + "type": "numeric_check", + "target_key": "suppression_ratio", + "weight": 70, + "params": { + "gt_value": 0.9995425785847105, + "tolerance": 0.09995425785847106 + } + }, + { + "metric_id": "suppression_success_accuracy", + "type": "categorical_check", + "target_key": "suppressed", + "weight": 30, + "params": { + "gt_value": false + } + } + ] + } +} diff --git a/foundational_analysis/cases/case26/case26_01.json b/foundational_analysis/cases/case26/case26_01.json new file mode 100644 index 0000000000000000000000000000000000000000..c5ef885acb3458d3104b7b42cf3ae95cda9977d2 --- /dev/null +++ b/foundational_analysis/cases/case26/case26_01.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -7.818929944225001, + "tolerance": 0.7818929944225002 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_02.json b/foundational_analysis/cases/case26/case26_02.json new file mode 100644 index 0000000000000000000000000000000000000000..83616552ef970347276bbaa7057bf8c63c6c0663 --- /dev/null +++ b/foundational_analysis/cases/case26/case26_02.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": 0.4710004534935912, + "tolerance": 0.047100045349359125 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "equal" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_03.json b/foundational_analysis/cases/case26/case26_03.json new file mode 100644 index 0000000000000000000000000000000000000000..114974a27b8b96f5184f8be99565cfbfa6e9b89a --- /dev/null +++ b/foundational_analysis/cases/case26/case26_03.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -7.2064163860920285, + "tolerance": 0.7206416386092029 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_04.json b/foundational_analysis/cases/case26/case26_04.json new file mode 100644 index 0000000000000000000000000000000000000000..02828d76dd2feb2011c43ec032ca8847dd045811 --- /dev/null +++ b/foundational_analysis/cases/case26/case26_04.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -0.9716555142339756, + "tolerance": 0.09716555142339756 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_05.json b/foundational_analysis/cases/case26/case26_05.json new file mode 100644 index 0000000000000000000000000000000000000000..1becfa9c5038a2ccbc79796029c946f2654b32c3 --- /dev/null +++ b/foundational_analysis/cases/case26/case26_05.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -0.3930382883416135, + "tolerance": 0.03930382883416135 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "equal" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_06.json b/foundational_analysis/cases/case26/case26_06.json new file mode 100644 index 0000000000000000000000000000000000000000..4c16d6d9e6b31a82991fe687c7b9b6d30a00a058 --- /dev/null +++ b/foundational_analysis/cases/case26/case26_06.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -2.0957526407379774, + "tolerance": 0.20957526407379776 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_07.json b/foundational_analysis/cases/case26/case26_07.json new file mode 100644 index 0000000000000000000000000000000000000000..3e46708a3425682164408a699b997fb542d7b2b7 --- /dev/null +++ b/foundational_analysis/cases/case26/case26_07.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": 0.5460304377902165, + "tolerance": 0.05460304377902166 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "frontal" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_08.json b/foundational_analysis/cases/case26/case26_08.json new file mode 100644 index 0000000000000000000000000000000000000000..8d87764a52d0bf6655cf83501845bbacfcc82de4 --- /dev/null +++ b/foundational_analysis/cases/case26/case26_08.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -0.6722551011805429, + "tolerance": 0.06722551011805429 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_09.json b/foundational_analysis/cases/case26/case26_09.json new file mode 100644 index 0000000000000000000000000000000000000000..9933a422dc10bbc01c2da3fafc19848bfebe00c0 --- /dev/null +++ b/foundational_analysis/cases/case26/case26_09.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": 2.758276327773375, + "tolerance": 0.2758276327773375 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "frontal" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_10.json b/foundational_analysis/cases/case26/case26_10.json new file mode 100644 index 0000000000000000000000000000000000000000..5960338818dacf79fec501d8f21dd60f3aa63bbd --- /dev/null +++ b/foundational_analysis/cases/case26/case26_10.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -2.014333949489119, + "tolerance": 0.2014333949489119 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_11.json b/foundational_analysis/cases/case26/case26_11.json new file mode 100644 index 0000000000000000000000000000000000000000..d14b8fc2442a5c676f67aa54ae4a44529868d667 --- /dev/null +++ b/foundational_analysis/cases/case26/case26_11.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -3.8428948444380087, + "tolerance": 0.3842894844438009 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_12.json b/foundational_analysis/cases/case26/case26_12.json new file mode 100644 index 0000000000000000000000000000000000000000..3242c89bdd228913090bd5c2e47c7aa082c51b24 --- /dev/null +++ b/foundational_analysis/cases/case26/case26_12.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -1.8329973027099475, + "tolerance": 0.18329973027099478 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_13.json b/foundational_analysis/cases/case26/case26_13.json new file mode 100644 index 0000000000000000000000000000000000000000..e3b8e52c6aedd589089a7d3544d3db2a5356c36e --- /dev/null +++ b/foundational_analysis/cases/case26/case26_13.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -3.7931197920508835, + "tolerance": 0.37931197920508836 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_14.json b/foundational_analysis/cases/case26/case26_14.json new file mode 100644 index 0000000000000000000000000000000000000000..c75e4547376fa54d4b51e1599effbbf6097a0c5c --- /dev/null +++ b/foundational_analysis/cases/case26/case26_14.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -5.807379344099614, + "tolerance": 0.5807379344099614 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_15.json b/foundational_analysis/cases/case26/case26_15.json new file mode 100644 index 0000000000000000000000000000000000000000..3680ac31068f9d48a3a107f3e38798d913dc6dc0 --- /dev/null +++ b/foundational_analysis/cases/case26/case26_15.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -8.70958830067044, + "tolerance": 0.870958830067044 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_16.json b/foundational_analysis/cases/case26/case26_16.json new file mode 100644 index 0000000000000000000000000000000000000000..045e68856c361b7a555be7d6982ddf420f5dc918 --- /dev/null +++ b/foundational_analysis/cases/case26/case26_16.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -31.89134955352457, + "tolerance": 3.1891349553524573 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_17.json b/foundational_analysis/cases/case26/case26_17.json new file mode 100644 index 0000000000000000000000000000000000000000..af137fdc6ded091edb763fe90a680083feae1522 --- /dev/null +++ b/foundational_analysis/cases/case26/case26_17.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -44.21023608904176, + "tolerance": 4.421023608904176 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_18.json b/foundational_analysis/cases/case26/case26_18.json new file mode 100644 index 0000000000000000000000000000000000000000..54dad04a39f4bef0f9b519a7e99ed08e671fd8c9 --- /dev/null +++ b/foundational_analysis/cases/case26/case26_18.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -17.730088990439747, + "tolerance": 1.7730088990439747 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_19.json b/foundational_analysis/cases/case26/case26_19.json new file mode 100644 index 0000000000000000000000000000000000000000..29a756521a288b97f75e897efe9106eb353e57a6 --- /dev/null +++ b/foundational_analysis/cases/case26/case26_19.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -30.12280978631655, + "tolerance": 3.0122809786316553 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_20.json b/foundational_analysis/cases/case26/case26_20.json new file mode 100644 index 0000000000000000000000000000000000000000..b5a06b6b8592af9645b668fb458ce07d7d4d993e --- /dev/null +++ b/foundational_analysis/cases/case26/case26_20.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -22.60451483410502, + "tolerance": 2.260451483410502 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_21.json b/foundational_analysis/cases/case26/case26_21.json new file mode 100644 index 0000000000000000000000000000000000000000..a67ad06f0cc28e5865014f8ae423662157fa4a0b --- /dev/null +++ b/foundational_analysis/cases/case26/case26_21.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -22.758431152479623, + "tolerance": 2.2758431152479623 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_22.json b/foundational_analysis/cases/case26/case26_22.json new file mode 100644 index 0000000000000000000000000000000000000000..2058d4a5ad347907c4855700ad83ea9957c93921 --- /dev/null +++ b/foundational_analysis/cases/case26/case26_22.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -10.159266061601379, + "tolerance": 1.015926606160138 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_23.json b/foundational_analysis/cases/case26/case26_23.json new file mode 100644 index 0000000000000000000000000000000000000000..8f8a1c1ab62be9fa8d9acb11545ce9a4199e584b --- /dev/null +++ b/foundational_analysis/cases/case26/case26_23.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -17.938680220599384, + "tolerance": 1.7938680220599386 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_24.json b/foundational_analysis/cases/case26/case26_24.json new file mode 100644 index 0000000000000000000000000000000000000000..37ad8e44b73d1a6c00b041fdfa73eb47c5d54da7 --- /dev/null +++ b/foundational_analysis/cases/case26/case26_24.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -26.602596239489664, + "tolerance": 2.6602596239489666 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case26/case26_25.json b/foundational_analysis/cases/case26/case26_25.json new file mode 100644 index 0000000000000000000000000000000000000000..464e1e05b55ad4e6d8d1cb7c897331ccb6c3b63c --- /dev/null +++ b/foundational_analysis/cases/case26/case26_25.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please compare the frontal-region and occipital-region alpha relative power ratios, report the difference (frontal minus occipital), and determine which region shows stronger alpha activity. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. If the absolute frontal-minus-occipital difference is less than 0.5 percentage points, report equal." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) frontal-occipital alpha relative power difference\n2) which region shows stronger alpha activity\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys in JSON must be EXACTLY \"alpha_ratio_diff\" and \"stronger_region\".\n5. \"alpha_ratio_diff\" must be a FLOAT or null.\n6. \"stronger_region\" must be one of \"frontal\", \"occipital\", \"equal\", or null.\n\n### OUTPUT TEMPLATE\n{\"alpha_ratio_diff\": , \"stronger_region\": }", + "metrics": [ + { + "metric_id": "alpha_ratio_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_ratio_diff", + "weight": 70, + "params": { + "gt_value": -18.37245675475863, + "tolerance": 1.837245675475863 + } + }, + { + "metric_id": "stronger_region_accuracy", + "type": "categorical_check", + "target_key": "stronger_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case27/case27_01.json b/foundational_analysis/cases/case27/case27_01.json new file mode 100644 index 0000000000000000000000000000000000000000..805b8eb4da8e6905ada4cdd77e80e4c07b35d648 --- /dev/null +++ b/foundational_analysis/cases/case27/case27_01.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please compare the strongest brain region in the first 10 minutes and the 10-20 minute segment using delta-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "frontal" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "frontal" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is frontal. The true strongest region in window 2 is frontal. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_02.json b/foundational_analysis/cases/case27/case27_02.json new file mode 100644 index 0000000000000000000000000000000000000000..3f75d7f24c6a29d5c703af636f0972720d33bf90 --- /dev/null +++ b/foundational_analysis/cases/case27/case27_02.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please compare the strongest brain region in the first 10 minutes and the 10-20 minute segment using delta-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "frontal" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is occipital. The true strongest region in window 2 is frontal. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: occipital to frontal. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_03.json b/foundational_analysis/cases/case27/case27_03.json new file mode 100644 index 0000000000000000000000000000000000000000..8a0fd1b08a4dd2b4ea494f995e7598745aa5ef96 --- /dev/null +++ b/foundational_analysis/cases/case27/case27_03.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please compare the strongest brain region in the first 10 minutes and the 10-20 minute segment using delta-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "frontal" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "frontal" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is frontal. The true strongest region in window 2 is frontal. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_04.json b/foundational_analysis/cases/case27/case27_04.json new file mode 100644 index 0000000000000000000000000000000000000000..8465f22762a327151a5b4974acf876f4290e20fd --- /dev/null +++ b/foundational_analysis/cases/case27/case27_04.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please compare the strongest brain region in the first 10 minutes and the 10-20 minute segment using delta-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "central" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "frontal" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is central. The true strongest region in window 2 is frontal. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: central to frontal. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_05.json b/foundational_analysis/cases/case27/case27_05.json new file mode 100644 index 0000000000000000000000000000000000000000..0c8c5d53d6142bad4b7dbe56e86232a506afd6ed --- /dev/null +++ b/foundational_analysis/cases/case27/case27_05.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please compare the strongest brain region in the first 10 minutes and the 10-20 minute segment using delta-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "frontal" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is occipital. The true strongest region in window 2 is frontal. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: occipital to frontal. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_06.json b/foundational_analysis/cases/case27/case27_06.json new file mode 100644 index 0000000000000000000000000000000000000000..7a19720974f627cee8748c1093757966d4f8a75f --- /dev/null +++ b/foundational_analysis/cases/case27/case27_06.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "frontal" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "frontal" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is frontal. The true strongest region in window 2 is frontal. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_07.json b/foundational_analysis/cases/case27/case27_07.json new file mode 100644 index 0000000000000000000000000000000000000000..578d7d9bb64b8523395c2f649167fc7294cf88af --- /dev/null +++ b/foundational_analysis/cases/case27/case27_07.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "parietal" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "parietal" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is parietal. The true strongest region in window 2 is parietal. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_08.json b/foundational_analysis/cases/case27/case27_08.json new file mode 100644 index 0000000000000000000000000000000000000000..cbcec937f9cf2a913a42894b213088269453ed47 --- /dev/null +++ b/foundational_analysis/cases/case27/case27_08.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "frontal" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "frontal" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is frontal. The true strongest region in window 2 is frontal. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_09.json b/foundational_analysis/cases/case27/case27_09.json new file mode 100644 index 0000000000000000000000000000000000000000..60408a3eed01b6fc9a61a6fa0792605fad32d959 --- /dev/null +++ b/foundational_analysis/cases/case27/case27_09.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "central" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "central" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is central. The true strongest region in window 2 is central. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_10.json b/foundational_analysis/cases/case27/case27_10.json new file mode 100644 index 0000000000000000000000000000000000000000..60f5f78161c839e4d015747996284e5eab820afc --- /dev/null +++ b/foundational_analysis/cases/case27/case27_10.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "parietal" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is parietal. The true strongest region in window 2 is occipital. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: parietal to occipital. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_11.json b/foundational_analysis/cases/case27/case27_11.json new file mode 100644 index 0000000000000000000000000000000000000000..9a15163555e85a36394924da2bd63032ef445488 --- /dev/null +++ b/foundational_analysis/cases/case27/case27_11.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is occipital. The true strongest region in window 2 is occipital. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_12.json b/foundational_analysis/cases/case27/case27_12.json new file mode 100644 index 0000000000000000000000000000000000000000..e89a36d26dc8c52a366e0b430378a46bc585edeb --- /dev/null +++ b/foundational_analysis/cases/case27/case27_12.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "parietal" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "parietal" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is parietal. The true strongest region in window 2 is parietal. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_13.json b/foundational_analysis/cases/case27/case27_13.json new file mode 100644 index 0000000000000000000000000000000000000000..c7ff6da7ced54564e2f6c94ad4e0ce1dfeb6134a --- /dev/null +++ b/foundational_analysis/cases/case27/case27_13.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is occipital. The true strongest region in window 2 is occipital. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_14.json b/foundational_analysis/cases/case27/case27_14.json new file mode 100644 index 0000000000000000000000000000000000000000..1e4b0fa3ad206f9638a716aca53704a3341c9bb7 --- /dev/null +++ b/foundational_analysis/cases/case27/case27_14.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "parietal" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "parietal" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is parietal. The true strongest region in window 2 is parietal. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_15.json b/foundational_analysis/cases/case27/case27_15.json new file mode 100644 index 0000000000000000000000000000000000000000..8287c2339f93189db10524f0f62385bc7c2a33cd --- /dev/null +++ b/foundational_analysis/cases/case27/case27_15.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is occipital. The true strongest region in window 2 is occipital. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_16.json b/foundational_analysis/cases/case27/case27_16.json new file mode 100644 index 0000000000000000000000000000000000000000..9ca09421229341b438b9c8bde75ebf68f64c568a --- /dev/null +++ b/foundational_analysis/cases/case27/case27_16.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "parietal" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is parietal. The true strongest region in window 2 is occipital. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: parietal to occipital. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_17.json b/foundational_analysis/cases/case27/case27_17.json new file mode 100644 index 0000000000000000000000000000000000000000..39d405552471a0cd31266c22d70e9c2bfdb9e970 --- /dev/null +++ b/foundational_analysis/cases/case27/case27_17.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is occipital. The true strongest region in window 2 is occipital. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_18.json b/foundational_analysis/cases/case27/case27_18.json new file mode 100644 index 0000000000000000000000000000000000000000..8c18d282a14674473fbc428be947519b26ea57fd --- /dev/null +++ b/foundational_analysis/cases/case27/case27_18.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is occipital. The true strongest region in window 2 is occipital. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_19.json b/foundational_analysis/cases/case27/case27_19.json new file mode 100644 index 0000000000000000000000000000000000000000..d2e467d9ed57953394161507c822e35bc14fb591 --- /dev/null +++ b/foundational_analysis/cases/case27/case27_19.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is occipital. The true strongest region in window 2 is occipital. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_20.json b/foundational_analysis/cases/case27/case27_20.json new file mode 100644 index 0000000000000000000000000000000000000000..9978398f1b618d3851857f1814c5cf14497733f8 --- /dev/null +++ b/foundational_analysis/cases/case27/case27_20.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is occipital. The true strongest region in window 2 is occipital. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_21.json b/foundational_analysis/cases/case27/case27_21.json new file mode 100644 index 0000000000000000000000000000000000000000..ce9f82dc26ce1e3b3c5e8881dd9cb570e23982ca --- /dev/null +++ b/foundational_analysis/cases/case27/case27_21.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is occipital. The true strongest region in window 2 is occipital. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_22.json b/foundational_analysis/cases/case27/case27_22.json new file mode 100644 index 0000000000000000000000000000000000000000..ff2982df52698736ccfadc8f98f0e3713e4bf84a --- /dev/null +++ b/foundational_analysis/cases/case27/case27_22.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is occipital. The true strongest region in window 2 is occipital. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_23.json b/foundational_analysis/cases/case27/case27_23.json new file mode 100644 index 0000000000000000000000000000000000000000..1c6166f2747d05616b1a64cece13b3493a00e563 --- /dev/null +++ b/foundational_analysis/cases/case27/case27_23.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is occipital. The true strongest region in window 2 is occipital. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_24.json b/foundational_analysis/cases/case27/case27_24.json new file mode 100644 index 0000000000000000000000000000000000000000..0bf69c28303b38544c31378751a77be2f0a40d7a --- /dev/null +++ b/foundational_analysis/cases/case27/case27_24.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is occipital. The true strongest region in window 2 is occipital. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case27/case27_25.json b/foundational_analysis/cases/case27/case27_25.json new file mode 100644 index 0000000000000000000000000000000000000000..4292c0bcfedeedbf7f68db73070546fdfb9161c7 --- /dev/null +++ b/foundational_analysis/cases/case27/case27_25.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please compare the strongest brain region in the first two 30-second windows using alpha-band relative energy across frontal, central, parietal, and occipital regions. Report the strongest region in each window and describe the region-state transition process. If the strongest region does not change, state that the status is stable. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract three outputs from the agent report:\n1) strongest region in window 1\n2) strongest region in window 2\n3) transition process description\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"window_1_strongest_region\", \"window_2_strongest_region\", and \"transition_process\".\n5. The first two values must be one of: frontal, central, parietal, occipital, or null.\n6. \"transition_process\" must be a string or null.\n\n### OUTPUT TEMPLATE\n{\"window_1_strongest_region\": , \"window_2_strongest_region\": , \"transition_process\": }", + "metrics": [ + { + "metric_id": "window_1_region_accuracy", + "type": "categorical_check", + "target_key": "window_1_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "window_2_region_accuracy", + "type": "categorical_check", + "target_key": "window_2_strongest_region", + "weight": 30, + "params": { + "gt_value": "occipital" + } + }, + { + "metric_id": "transition_process_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task that compares the strongest brain region across two time windows. The true strongest region in window 1 is occipital. The true strongest region in window 2 is occipital. Mark the response as correct only if its transition narrative matches the real relationship between these two windows. If the strongest region is the same in both windows, the response must clearly indicate stability or no transition. If the strongest region changes, the response must clearly describe the shift from the first region to the second region. The expected transition pattern is: stable. If the response describes the wrong direction, invents a transition when the state is stable, or claims stability when a shift occurred, mark it as incorrect. Output only a valid JSON object. The JSON must contain exactly one key named status. The value of status must be either True or False. Do not output Markdown, explanations, additional fields, or any natural language." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_01.json b/foundational_analysis/cases/case28/case28_01.json new file mode 100644 index 0000000000000000000000000000000000000000..f60e73b7b26f506bb777c7927233128c14b5bbb0 --- /dev/null +++ b/foundational_analysis/cases/case28/case28_01.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "occipital", + "central", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is occipital, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_02.json b/foundational_analysis/cases/case28/case28_02.json new file mode 100644 index 0000000000000000000000000000000000000000..6c53d585565e1984195407ae77753bebb7f6968d --- /dev/null +++ b/foundational_analysis/cases/case28/case28_02.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "frontal", + "occipital", + "central" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": false + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is frontal, and posterior-dominant alpha is absent. The explanation must make it clear that posterior regions are not strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_03.json b/foundational_analysis/cases/case28/case28_03.json new file mode 100644 index 0000000000000000000000000000000000000000..faa6fcfcf78e4af2d8e851aa8cd542191905820a --- /dev/null +++ b/foundational_analysis/cases/case28/case28_03.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "occipital", + "central", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is occipital, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_04.json b/foundational_analysis/cases/case28/case28_04.json new file mode 100644 index 0000000000000000000000000000000000000000..92f1c4de5023a76b5978b2971eaa631340b5fb7d --- /dev/null +++ b/foundational_analysis/cases/case28/case28_04.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "occipital", + "central", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is occipital, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_05.json b/foundational_analysis/cases/case28/case28_05.json new file mode 100644 index 0000000000000000000000000000000000000000..a5640f5a3e92e6fb4cfcc9e30b1746f40fca0fce --- /dev/null +++ b/foundational_analysis/cases/case28/case28_05.json @@ -0,0 +1,49 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "occipital", + "central", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is occipital, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_06.json b/foundational_analysis/cases/case28/case28_06.json new file mode 100644 index 0000000000000000000000000000000000000000..130f63ac08bd3f116389e2ffa0666d4c422b4aed --- /dev/null +++ b/foundational_analysis/cases/case28/case28_06.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "occipital", + "parietal", + "central", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is occipital, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_07.json b/foundational_analysis/cases/case28/case28_07.json new file mode 100644 index 0000000000000000000000000000000000000000..847d79af68cb47c3007694e324504e87bdf97be7 --- /dev/null +++ b/foundational_analysis/cases/case28/case28_07.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "parietal", + "frontal", + "central", + "occipital" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is parietal, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_08.json b/foundational_analysis/cases/case28/case28_08.json new file mode 100644 index 0000000000000000000000000000000000000000..bdd0ea20ab8419b39d457d896892c5204d4101eb --- /dev/null +++ b/foundational_analysis/cases/case28/case28_08.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "occipital", + "central", + "parietal", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is occipital, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_09.json b/foundational_analysis/cases/case28/case28_09.json new file mode 100644 index 0000000000000000000000000000000000000000..ad48c17763751d46b7916d4e4e8da5f60792676e --- /dev/null +++ b/foundational_analysis/cases/case28/case28_09.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "central", + "frontal", + "parietal", + "occipital" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": false + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is central, and posterior-dominant alpha is absent. The explanation must make it clear that posterior regions are not strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_10.json b/foundational_analysis/cases/case28/case28_10.json new file mode 100644 index 0000000000000000000000000000000000000000..e3722e5231bff8531a36892e0ebe24ce76ef5c0a --- /dev/null +++ b/foundational_analysis/cases/case28/case28_10.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "occipital", + "central", + "parietal", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is occipital, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_11.json b/foundational_analysis/cases/case28/case28_11.json new file mode 100644 index 0000000000000000000000000000000000000000..13b7f83f125cce3773208b856d5fffa089aaf4f5 --- /dev/null +++ b/foundational_analysis/cases/case28/case28_11.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "parietal", + "occipital", + "central", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is parietal, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_12.json b/foundational_analysis/cases/case28/case28_12.json new file mode 100644 index 0000000000000000000000000000000000000000..43119e8f117304774d00cf8d44f400b970d2e50c --- /dev/null +++ b/foundational_analysis/cases/case28/case28_12.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "parietal", + "central", + "occipital", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is parietal, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_13.json b/foundational_analysis/cases/case28/case28_13.json new file mode 100644 index 0000000000000000000000000000000000000000..ca0c5da965a19e24259e03986281818a3a95e4cb --- /dev/null +++ b/foundational_analysis/cases/case28/case28_13.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "occipital", + "parietal", + "central", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is occipital, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_14.json b/foundational_analysis/cases/case28/case28_14.json new file mode 100644 index 0000000000000000000000000000000000000000..e9e5d89603f8ab1d757c692125bd3bbfc96ec12d --- /dev/null +++ b/foundational_analysis/cases/case28/case28_14.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "parietal", + "occipital", + "central", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is parietal, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_15.json b/foundational_analysis/cases/case28/case28_15.json new file mode 100644 index 0000000000000000000000000000000000000000..71ff8af1575450ce1e1eaa0520ca8aff4f4597b9 --- /dev/null +++ b/foundational_analysis/cases/case28/case28_15.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "occipital", + "parietal", + "central", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is occipital, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_16.json b/foundational_analysis/cases/case28/case28_16.json new file mode 100644 index 0000000000000000000000000000000000000000..7c6f0eb4b2b281ffa80e1150ed3a6a5be9f7e96d --- /dev/null +++ b/foundational_analysis/cases/case28/case28_16.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "parietal", + "occipital", + "central", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is parietal, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_17.json b/foundational_analysis/cases/case28/case28_17.json new file mode 100644 index 0000000000000000000000000000000000000000..eda2698cb2b1c5e8debc5444b0c3384e22f7a3c7 --- /dev/null +++ b/foundational_analysis/cases/case28/case28_17.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "occipital", + "parietal", + "central", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is occipital, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_18.json b/foundational_analysis/cases/case28/case28_18.json new file mode 100644 index 0000000000000000000000000000000000000000..4d5afc74fd09d7627689eb562e203b1b5f84044a --- /dev/null +++ b/foundational_analysis/cases/case28/case28_18.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "occipital", + "parietal", + "central", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is occipital, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_19.json b/foundational_analysis/cases/case28/case28_19.json new file mode 100644 index 0000000000000000000000000000000000000000..578ba21fb053a46e61d091e6d4157ba17c8de1ce --- /dev/null +++ b/foundational_analysis/cases/case28/case28_19.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "occipital", + "parietal", + "central", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is occipital, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_20.json b/foundational_analysis/cases/case28/case28_20.json new file mode 100644 index 0000000000000000000000000000000000000000..780fbf8fc0ceeae97856a441a2d6997046089dff --- /dev/null +++ b/foundational_analysis/cases/case28/case28_20.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "occipital", + "parietal", + "central", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is occipital, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_21.json b/foundational_analysis/cases/case28/case28_21.json new file mode 100644 index 0000000000000000000000000000000000000000..868af787187ab69d972b6c1f93000923f4ff819a --- /dev/null +++ b/foundational_analysis/cases/case28/case28_21.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "occipital", + "parietal", + "central", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is occipital, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_22.json b/foundational_analysis/cases/case28/case28_22.json new file mode 100644 index 0000000000000000000000000000000000000000..677f06bb1c5572a3921818e18f2969d968cc01c8 --- /dev/null +++ b/foundational_analysis/cases/case28/case28_22.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "occipital", + "parietal", + "frontal", + "central" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is occipital, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_23.json b/foundational_analysis/cases/case28/case28_23.json new file mode 100644 index 0000000000000000000000000000000000000000..94029f8ec3a1f06bd2ba159bcf08d85d4576fcee --- /dev/null +++ b/foundational_analysis/cases/case28/case28_23.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "occipital", + "parietal", + "central", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is occipital, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_24.json b/foundational_analysis/cases/case28/case28_24.json new file mode 100644 index 0000000000000000000000000000000000000000..ef7d17de93bd78ccc2f17886129ed9d1715148b5 --- /dev/null +++ b/foundational_analysis/cases/case28/case28_24.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "occipital", + "parietal", + "central", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is occipital, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case28/case28_25.json b/foundational_analysis/cases/case28/case28_25.json new file mode 100644 index 0000000000000000000000000000000000000000..ec507c7ab1b8561187d7cb29ba9c2228b3fbe6c4 --- /dev/null +++ b/foundational_analysis/cases/case28/case28_25.json @@ -0,0 +1,50 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please compute the alpha relative power ratio for each brain region, output the brain-region names in descending order of alpha relative power, determine whether posterior-dominant alpha exists, and explain your conclusion. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Compute relative power separately for each channel, and define each region's relative-power score as the arithmetic mean of the channel-wise relative-power values within that region." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) region alpha ranking in descending order\n2) whether posterior-dominant alpha exists\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"region_alpha_rank\" and \"posterior_dominant_alpha\".\n5. For \"region_alpha_rank\", read the final descending region ranking and keep only exact canonical regions from: frontal, central, parietal, occipital.\n6. Lowercase every retained region, remove duplicates, and preserve their relative order. Ignore unsupported labels such as temporal, mastoid, mixed, or compound labels such as right temporal/parietal; do not map or guess them.\n7. Example: [Central, Frontal, Occipital, Right Temporal/Parietal] -> [\"central\",\"frontal\",\"occipital\"].\n8. Map an explicit final YES/present/exists/true conclusion to true and an explicit NO/absent/does not exist/false conclusion to false. If statements conflict, use only the final conclusion.\n9. If no canonical ranked region is identifiable, set region_alpha_rank to null. If the posterior-dominance conclusion is unclear, set posterior_dominant_alpha to null. Do not guess.\n\n### OUTPUT TEMPLATE\n{\"region_alpha_rank\": [\"region1\",\"region2\"], \"posterior_dominant_alpha\": }", + "metrics": [ + { + "metric_id": "region_alpha_rank_accuracy", + "type": "sequence_match_check", + "target_key": "region_alpha_rank", + "weight": 40, + "params": { + "gt_value": [ + "occipital", + "parietal", + "central", + "frontal" + ], + "match_mode": "exact_order" + } + }, + { + "metric_id": "posterior_dominant_alpha_accuracy", + "type": "categorical_check", + "target_key": "posterior_dominant_alpha", + "weight": 30, + "params": { + "gt_value": true + } + }, + { + "metric_id": "posterior_alpha_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 30, + "params": { + "judge_prompt": "Judge whether the explanation is logically consistent with the true region ranking and the posterior-dominant alpha label. The true top-ranked region is occipital, and posterior-dominant alpha is present. The explanation must make it clear that a posterior region (parietal or occipital) is strongest. If the explanation contradicts the true ranking or the posterior-dominant alpha label, mark it as incorrect. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_01.json b/foundational_analysis/cases/case29/case29_01.json new file mode 100644 index 0000000000000000000000000000000000000000..d11eae7d8ff99a42cb210b9d67ee09947a250dab --- /dev/null +++ b/foundational_analysis/cases/case29/case29_01.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.35175327116537425, + "tolerance": 0.017587663558268712 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "C4-F4", + "C3-F3", + "C4-O2" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_02.json b/foundational_analysis/cases/case29/case29_02.json new file mode 100644 index 0000000000000000000000000000000000000000..4979bd7428e8f665110b8ae5890b6b86d1ed14ba --- /dev/null +++ b/foundational_analysis/cases/case29/case29_02.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.41326966922760966, + "tolerance": 0.020663483461380486 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "C4-F4", + "C3-F3", + "C3-C4" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_03.json b/foundational_analysis/cases/case29/case29_03.json new file mode 100644 index 0000000000000000000000000000000000000000..68459ffd35359caea39d3f4de86d4873277c2349 --- /dev/null +++ b/foundational_analysis/cases/case29/case29_03.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.2999536044430944, + "tolerance": 0.01499768022215472 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "C4-F4", + "C3-F3", + "O1-O2" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_04.json b/foundational_analysis/cases/case29/case29_04.json new file mode 100644 index 0000000000000000000000000000000000000000..26590680c6917240dc35ada3ded55f0aeb1cc32f --- /dev/null +++ b/foundational_analysis/cases/case29/case29_04.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.28814339260657307, + "tolerance": 0.014407169630328655 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "C4-F4", + "C3-F3", + "C3-O1" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_05.json b/foundational_analysis/cases/case29/case29_05.json new file mode 100644 index 0000000000000000000000000000000000000000..2d38b125ac109cf41fecb49ead50881b7524d9b8 --- /dev/null +++ b/foundational_analysis/cases/case29/case29_05.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.30566237362716253, + "tolerance": 0.015283118681358128 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "C4-F4", + "C3-F3", + "C4-O2" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_06.json b/foundational_analysis/cases/case29/case29_06.json new file mode 100644 index 0000000000000000000000000000000000000000..7187c939a1d7239348a17c88174532285fe2d769 --- /dev/null +++ b/foundational_analysis/cases/case29/case29_06.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.41587174007641675, + "tolerance": 0.02079358700382084 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "P3-P5", + "CPZ-PZ", + "P6-PO4" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_07.json b/foundational_analysis/cases/case29/case29_07.json new file mode 100644 index 0000000000000000000000000000000000000000..65b1595f44453e8df9af45e22c9376c364a6d48c --- /dev/null +++ b/foundational_analysis/cases/case29/case29_07.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.36480923203388194, + "tolerance": 0.018240461601694097 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "T8-TP10", + "FT8-TP10", + "C6-F3" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_08.json b/foundational_analysis/cases/case29/case29_08.json new file mode 100644 index 0000000000000000000000000000000000000000..c7d0f6c251b2e8dc26e33c4f5074d00bbd8d5d12 --- /dev/null +++ b/foundational_analysis/cases/case29/case29_08.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.41539431245881714, + "tolerance": 0.02076971562294086 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "F5-F7", + "AF7-F5", + "AF7-F7" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_09.json b/foundational_analysis/cases/case29/case29_09.json new file mode 100644 index 0000000000000000000000000000000000000000..9c5f74de49c4841a6d5bd018d7ec8f9f50700b1b --- /dev/null +++ b/foundational_analysis/cases/case29/case29_09.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.4708000502146488, + "tolerance": 0.023540002510732443 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "F7-FT7", + "F4-F6", + "F3-F5" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_10.json b/foundational_analysis/cases/case29/case29_10.json new file mode 100644 index 0000000000000000000000000000000000000000..5b8a8e1269332aa612c6a5d77eba9bace7d3fb42 --- /dev/null +++ b/foundational_analysis/cases/case29/case29_10.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.47119803830722307, + "tolerance": 0.023559901915361156 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "P1-P4", + "P1-P5", + "CP3-P5" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_11.json b/foundational_analysis/cases/case29/case29_11.json new file mode 100644 index 0000000000000000000000000000000000000000..0476a95598ada5516262befee2d45f6dacb06d2d --- /dev/null +++ b/foundational_analysis/cases/case29/case29_11.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.3375081523984955, + "tolerance": 0.016875407619924777 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "PO6-PO8", + "P4-P6", + "PO5-PO7" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_12.json b/foundational_analysis/cases/case29/case29_12.json new file mode 100644 index 0000000000000000000000000000000000000000..eb53b17001cd45b85e0309f43bcbf714f8687662 --- /dev/null +++ b/foundational_analysis/cases/case29/case29_12.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.32206523430566286, + "tolerance": 0.016103261715283142 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "PO3-PO5", + "F4-F6", + "CP2-CP4" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_13.json b/foundational_analysis/cases/case29/case29_13.json new file mode 100644 index 0000000000000000000000000000000000000000..b1adb5f2ee4e72f3508600a216f4957dbba109e5 --- /dev/null +++ b/foundational_analysis/cases/case29/case29_13.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.3255120453571638, + "tolerance": 0.01627560226785819 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "PO5-PO7", + "PO4-PO6", + "P4-P6" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_14.json b/foundational_analysis/cases/case29/case29_14.json new file mode 100644 index 0000000000000000000000000000000000000000..d532a4d42d5a56dfd9d514380fe2e83e146ffe89 --- /dev/null +++ b/foundational_analysis/cases/case29/case29_14.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.35428962170927186, + "tolerance": 0.017714481085463595 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "O2-PO4", + "PO6-PO8", + "P4-P6" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_15.json b/foundational_analysis/cases/case29/case29_15.json new file mode 100644 index 0000000000000000000000000000000000000000..ccd71a0435493060f71c27b8aa273f7fe0c51fab --- /dev/null +++ b/foundational_analysis/cases/case29/case29_15.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.3865754924113744, + "tolerance": 0.01932877462056872 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "P4-P6", + "P8-PO8", + "P7-PO7" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_16.json b/foundational_analysis/cases/case29/case29_16.json new file mode 100644 index 0000000000000000000000000000000000000000..491cf4bddeb5b9e6a56762e3c5ffaec44f33870f --- /dev/null +++ b/foundational_analysis/cases/case29/case29_16.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.33651670441588755, + "tolerance": 0.016825835220794377 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "F4-FZ", + "FP1-FP2", + "F3-FZ" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_17.json b/foundational_analysis/cases/case29/case29_17.json new file mode 100644 index 0000000000000000000000000000000000000000..24a5414696f137b0c68173e90cdb7576ed756790 --- /dev/null +++ b/foundational_analysis/cases/case29/case29_17.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.4508436014086369, + "tolerance": 0.022542180070431846 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "F4-FZ", + "FP1-FP2", + "F3-FZ" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_18.json b/foundational_analysis/cases/case29/case29_18.json new file mode 100644 index 0000000000000000000000000000000000000000..fcf2ecadbedddad9ffcffd2b912452a5659d8147 --- /dev/null +++ b/foundational_analysis/cases/case29/case29_18.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.5670255319571492, + "tolerance": 0.028351276597857463 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "F4-FZ", + "F3-FZ", + "F3-FP1" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_19.json b/foundational_analysis/cases/case29/case29_19.json new file mode 100644 index 0000000000000000000000000000000000000000..7901c5272013550e22e55f1549f7cad5462a6dea --- /dev/null +++ b/foundational_analysis/cases/case29/case29_19.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.34590071476014983, + "tolerance": 0.017295035738007492 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "FP1-FZ", + "F4-F8", + "F3-F7" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_20.json b/foundational_analysis/cases/case29/case29_20.json new file mode 100644 index 0000000000000000000000000000000000000000..e64f1fefcf0530334f6fa4d6a7a1a83664b7a70c --- /dev/null +++ b/foundational_analysis/cases/case29/case29_20.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.42717435948314564, + "tolerance": 0.021358717974157282 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "FP1-FP2", + "F3-FP1", + "F3-F7" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_21.json b/foundational_analysis/cases/case29/case29_21.json new file mode 100644 index 0000000000000000000000000000000000000000..b0f46fbc8ab9bd861c5147a2f5190ceb8343ebb2 --- /dev/null +++ b/foundational_analysis/cases/case29/case29_21.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.4260101199281751, + "tolerance": 0.02130050599640876 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "F4-FZ", + "FP1-FP2", + "F3-FP1" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_22.json b/foundational_analysis/cases/case29/case29_22.json new file mode 100644 index 0000000000000000000000000000000000000000..7bf6185e27c52ebc2628529f8636c66fa0128f3a --- /dev/null +++ b/foundational_analysis/cases/case29/case29_22.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.5101197395785643, + "tolerance": 0.025505986978928214 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "F7-FP1", + "FP1-FP2", + "C4-T8" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_23.json b/foundational_analysis/cases/case29/case29_23.json new file mode 100644 index 0000000000000000000000000000000000000000..596ee9f7fd313568e6e5df8226adecdf4526e4ef --- /dev/null +++ b/foundational_analysis/cases/case29/case29_23.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.41009133126738706, + "tolerance": 0.020504566563369353 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "C4-CZ", + "C3-CZ", + "P4-PZ" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_24.json b/foundational_analysis/cases/case29/case29_24.json new file mode 100644 index 0000000000000000000000000000000000000000..b67f99b9f8b75bad8a956d5b6e45a7ccb1eefdf0 --- /dev/null +++ b/foundational_analysis/cases/case29/case29_24.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.46269915554690816, + "tolerance": 0.023134957777345408 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "F4-FZ", + "F3-FZ", + "F4-FP2" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case29/case29_25.json b/foundational_analysis/cases/case29/case29_25.json new file mode 100644 index 0000000000000000000000000000000000000000..66730a55497e02c9ded6487e6da0ad0d971c517f --- /dev/null +++ b/foundational_analysis/cases/case29/case29_25.json @@ -0,0 +1,43 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please select standard EEG channels and apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals. Then use the first 60 seconds of the filtered data, extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, report the global mean PLV, and report the top-3 channel pairs ranked from highest to lowest PLV." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) global mean PLV\n2) top-3 channel pairs ranked by PLV\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"global_mean_plv\" and \"top3_plv_pairs\".\n5. \"global_mean_plv\" must be a FLOAT or null.\n6. \"top3_plv_pairs\" must be an array of up to 3 strings or null. Normalize EACH reported pair independently.\n7. First identify the two complete source channel labels; a referenced label such as F4-A1 is one complete channel label, and its hyphen must not be treated as the pair separator.\n8. For each source channel separately, uppercase it and remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG. Never output a reference electrode (A1/A2/LE/REF/M1/M2/AVG) as a pair member.\n9. Convert legacy aliases after removing the reference suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Sort the two normalized channel labels in ascending lexicographic order, then join them with exactly one hyphen using format \"CH1-CH2\" (uppercase, no spaces). If the report already uses a canonical pair such as F4-C4, normalize it to C4-F4. Each final label may contain only uppercase A-Z letters and digits; mixed-case forms such as Fp1 or Fp2 are forbidden and must become FP1 or FP2.\n11. Required examples: \"F4-A1 and C4-A1\" -> \"C4-F4\"; \"F4-C4\" -> \"C4-F4\"; \"T4-A1 and T3-A2\" -> \"T7-T8\"; \"Fp2 and Fp1\" -> \"FP1-FP2\".\n12. Preserve the pairs in descending PLV order as reported. Do not sort the array lexicographically. If any pair cannot be resolved into exactly two non-reference channels, set \"top3_plv_pairs\" to null; do not guess.\n\n### OUTPUT TEMPLATE\n{\"global_mean_plv\": , \"top3_plv_pairs\": [\"CH1-CH2\",\"CH3-CH4\",\"CH5-CH6\"]}", + "metrics": [ + { + "metric_id": "global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "global_mean_plv", + "weight": 40, + "params": { + "gt_value": 0.426363922399868, + "tolerance": 0.0213181961199934 + } + }, + { + "metric_id": "top3_plv_pair_set_accuracy", + "type": "set_match_check", + "target_key": "top3_plv_pairs", + "weight": 60, + "params": { + "gt_value": [ + "F8-FP2", + "F7-FP1", + "O1-PZ" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_01.json b/foundational_analysis/cases/case30/case30_01.json new file mode 100644 index 0000000000000000000000000000000000000000..e9941cab4eb4bb4a7aa428d6057701dcc237801b --- /dev/null +++ b/foundational_analysis/cases/case30/case30_01.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.33870515895882675, + "tolerance": 0.01693525794794134 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.34405984632756176, + "tolerance": 0.017202992316378087 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.3436785796980445, + "tolerance": 0.017183928984902227 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.33105312252380337, + "tolerance": 0.01655265612619017 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is theta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_02.json b/foundational_analysis/cases/case30/case30_02.json new file mode 100644 index 0000000000000000000000000000000000000000..43142042b4de6f43edecb03354a41566d014079b --- /dev/null +++ b/foundational_analysis/cases/case30/case30_02.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.4502311431466776, + "tolerance": 0.022511557157333884 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.4009626389840913, + "tolerance": 0.020048131949204566 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.40663128057569986, + "tolerance": 0.020331564028784995 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.35848312505214586, + "tolerance": 0.017924156252607293 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is delta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_03.json b/foundational_analysis/cases/case30/case30_03.json new file mode 100644 index 0000000000000000000000000000000000000000..7fb575d5d20882f4d5582a87a867a822456fcf61 --- /dev/null +++ b/foundational_analysis/cases/case30/case30_03.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.315135407020352, + "tolerance": 0.0157567703510176 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.3161678365193695, + "tolerance": 0.015808391825968476 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.29857636017372385, + "tolerance": 0.014928818008686194 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.30272094381981557, + "tolerance": 0.015136047190990778 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is theta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_04.json b/foundational_analysis/cases/case30/case30_04.json new file mode 100644 index 0000000000000000000000000000000000000000..efc932aee535d370e1c0c0018a43a99e39f7b52a --- /dev/null +++ b/foundational_analysis/cases/case30/case30_04.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.3574922971910697, + "tolerance": 0.017874614859553484 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.3225410816605111, + "tolerance": 0.016127054083025557 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.29275922852906583, + "tolerance": 0.014637961426453292 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.3057533460178835, + "tolerance": 0.015287667300894175 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is delta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_05.json b/foundational_analysis/cases/case30/case30_05.json new file mode 100644 index 0000000000000000000000000000000000000000..224efedc69dc2430868c6fa3ad70af8eca84b422 --- /dev/null +++ b/foundational_analysis/cases/case30/case30_05.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.37248488625217097, + "tolerance": 0.01862424431260855 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.3868350139655193, + "tolerance": 0.019341750698275966 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.30897264425832005, + "tolerance": 0.015448632212916003 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.2979966564621495, + "tolerance": 0.014899832823107476 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is theta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_06.json b/foundational_analysis/cases/case30/case30_06.json new file mode 100644 index 0000000000000000000000000000000000000000..57e19211de57261385604014c0ed997a24b599b1 --- /dev/null +++ b/foundational_analysis/cases/case30/case30_06.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.4175459308089152, + "tolerance": 0.02087729654044576 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.3708450458597539, + "tolerance": 0.018542252292987695 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.42444737248081127, + "tolerance": 0.021222368624040564 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.34715182687324275, + "tolerance": 0.01735759134366214 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is alpha. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_07.json b/foundational_analysis/cases/case30/case30_07.json new file mode 100644 index 0000000000000000000000000000000000000000..201d2f73f04d234bc46c5b6426b2876a900cae8b --- /dev/null +++ b/foundational_analysis/cases/case30/case30_07.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.38388616135399334, + "tolerance": 0.01919430806769967 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.5150973768718287, + "tolerance": 0.025754868843591433 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.37315650496528546, + "tolerance": 0.018657825248264273 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.3079125456610608, + "tolerance": 0.015395627283053041 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is theta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_08.json b/foundational_analysis/cases/case30/case30_08.json new file mode 100644 index 0000000000000000000000000000000000000000..0d90b70aedd5d54f7a2892d3eeb15aeac47a278b --- /dev/null +++ b/foundational_analysis/cases/case30/case30_08.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.48899136993105374, + "tolerance": 0.02444956849655269 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.4085409161012921, + "tolerance": 0.020427045805064607 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.39538112973814427, + "tolerance": 0.019769056486907216 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.35753909180281995, + "tolerance": 0.017876954590141 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is delta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_09.json b/foundational_analysis/cases/case30/case30_09.json new file mode 100644 index 0000000000000000000000000000000000000000..497f1af1088dab25858a8d1c30e0ab227d5fb9f1 --- /dev/null +++ b/foundational_analysis/cases/case30/case30_09.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.5807597445360296, + "tolerance": 0.02903798722680148 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.5301284390543806, + "tolerance": 0.02650642195271903 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.4737086613104401, + "tolerance": 0.023685433065522007 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.46748278496746465, + "tolerance": 0.023374139248373234 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is delta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_10.json b/foundational_analysis/cases/case30/case30_10.json new file mode 100644 index 0000000000000000000000000000000000000000..60aca91be90bc1bfbfd51c6889ae4c8dbd40bfae --- /dev/null +++ b/foundational_analysis/cases/case30/case30_10.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.5328131128544772, + "tolerance": 0.02664065564272386 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.5330488342680207, + "tolerance": 0.02665244171340104 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.47437804126270433, + "tolerance": 0.02371890206313522 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.38006049094222916, + "tolerance": 0.01900302454711146 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is theta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_11.json b/foundational_analysis/cases/case30/case30_11.json new file mode 100644 index 0000000000000000000000000000000000000000..cd486fbdfc7cbf952fc480b70fcd7669014804e2 --- /dev/null +++ b/foundational_analysis/cases/case30/case30_11.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.3601906928066973, + "tolerance": 0.018009534640334866 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.3435351503251644, + "tolerance": 0.01717675751625822 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.32752526148831435, + "tolerance": 0.01637626307441572 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.2249627264192843, + "tolerance": 0.011248136320964216 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is delta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_12.json b/foundational_analysis/cases/case30/case30_12.json new file mode 100644 index 0000000000000000000000000000000000000000..ed46f8493feeaa84ce995abd39d5bce877264485 --- /dev/null +++ b/foundational_analysis/cases/case30/case30_12.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.31769439412145845, + "tolerance": 0.015884719706072924 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.32533324405965947, + "tolerance": 0.016266662202982975 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.31677480733100466, + "tolerance": 0.015838740366550235 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.2656337040205575, + "tolerance": 0.013281685201027876 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is theta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_13.json b/foundational_analysis/cases/case30/case30_13.json new file mode 100644 index 0000000000000000000000000000000000000000..7baa55ec7c7db8e18b8c078bc7f240b9b1539207 --- /dev/null +++ b/foundational_analysis/cases/case30/case30_13.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.3719940148556445, + "tolerance": 0.018599700742782225 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.35978669688366965, + "tolerance": 0.017989334844183483 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.31912959125461066, + "tolerance": 0.015956479562730535 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.24347466887381405, + "tolerance": 0.012173733443690703 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is delta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_14.json b/foundational_analysis/cases/case30/case30_14.json new file mode 100644 index 0000000000000000000000000000000000000000..269cad12baa41e26d5b92346768a5d51ecd43a38 --- /dev/null +++ b/foundational_analysis/cases/case30/case30_14.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.32840606592418514, + "tolerance": 0.01642030329620926 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.36246552732614273, + "tolerance": 0.018123276366307138 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.3423757198555794, + "tolerance": 0.01711878599277897 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.2687322515514646, + "tolerance": 0.01343661257757323 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is theta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_15.json b/foundational_analysis/cases/case30/case30_15.json new file mode 100644 index 0000000000000000000000000000000000000000..338093e59dfee2d7c87ffae5c2c0e2745e70ba47 --- /dev/null +++ b/foundational_analysis/cases/case30/case30_15.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.3419320922196381, + "tolerance": 0.017096604610981905 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.37786376560045726, + "tolerance": 0.018893188280022864 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.3856198789876578, + "tolerance": 0.01928099394938289 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.33173298042359645, + "tolerance": 0.016586649021179824 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is alpha. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_16.json b/foundational_analysis/cases/case30/case30_16.json new file mode 100644 index 0000000000000000000000000000000000000000..b005715b173c7d071fc72118d04565428a9444db --- /dev/null +++ b/foundational_analysis/cases/case30/case30_16.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.4579894429490657, + "tolerance": 0.022899472147453286 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.3673106516860453, + "tolerance": 0.018365532584302267 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.33285170753633286, + "tolerance": 0.016642585376816645 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.3248234277937441, + "tolerance": 0.01624117138968721 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is delta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_17.json b/foundational_analysis/cases/case30/case30_17.json new file mode 100644 index 0000000000000000000000000000000000000000..d7e6c715308b2a26e4b989d7976a9b1a4687c88f --- /dev/null +++ b/foundational_analysis/cases/case30/case30_17.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.4362390486837362, + "tolerance": 0.02181195243418681 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.4200678566500757, + "tolerance": 0.021003392832503784 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.4301560344761331, + "tolerance": 0.021507801723806657 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.3537174269276258, + "tolerance": 0.01768587134638129 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is delta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_18.json b/foundational_analysis/cases/case30/case30_18.json new file mode 100644 index 0000000000000000000000000000000000000000..375ac028db9b792533467e48e8a05574933a3bec --- /dev/null +++ b/foundational_analysis/cases/case30/case30_18.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.5702406110239906, + "tolerance": 0.02851203055119953 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.5614991390991084, + "tolerance": 0.028074956954955422 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.5257437287919953, + "tolerance": 0.026287186439599763 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.44626730378707297, + "tolerance": 0.02231336518935365 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is delta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_19.json b/foundational_analysis/cases/case30/case30_19.json new file mode 100644 index 0000000000000000000000000000000000000000..b10b6c0b8c8177cdf50e8ddcb5ea2583e2861f27 --- /dev/null +++ b/foundational_analysis/cases/case30/case30_19.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.8571368830121525, + "tolerance": 0.04285684415060763 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.8740382667386682, + "tolerance": 0.043701913336933414 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.35102192096629675, + "tolerance": 0.017551096048314837 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.24711033067504426, + "tolerance": 0.012355516533752213 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is theta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_20.json b/foundational_analysis/cases/case30/case30_20.json new file mode 100644 index 0000000000000000000000000000000000000000..a5340ca66c348602fa4e2062818949f75124e9f6 --- /dev/null +++ b/foundational_analysis/cases/case30/case30_20.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.4001235994967141, + "tolerance": 0.020006179974835708 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.4045192830422137, + "tolerance": 0.020225964152110684 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.4223197445643653, + "tolerance": 0.021115987228218266 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.3225343407974873, + "tolerance": 0.016126717039874363 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is alpha. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_21.json b/foundational_analysis/cases/case30/case30_21.json new file mode 100644 index 0000000000000000000000000000000000000000..5943d776ad802e2e0494913901b53a18723a30c2 --- /dev/null +++ b/foundational_analysis/cases/case30/case30_21.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.418090107329654, + "tolerance": 0.020904505366482702 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.48211240966461893, + "tolerance": 0.02410562048323095 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.44363603957411885, + "tolerance": 0.022181801978705944 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.3512130468878125, + "tolerance": 0.017560652344390627 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is theta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_22.json b/foundational_analysis/cases/case30/case30_22.json new file mode 100644 index 0000000000000000000000000000000000000000..c98455b215ec8b9460414110eca99bbddd6c4cc8 --- /dev/null +++ b/foundational_analysis/cases/case30/case30_22.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.5915304172727218, + "tolerance": 0.02957652086363609 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.57606447173796, + "tolerance": 0.028803223586898005 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.5142643865008782, + "tolerance": 0.025713219325043914 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.505115395888964, + "tolerance": 0.025255769794448202 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is delta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_23.json b/foundational_analysis/cases/case30/case30_23.json new file mode 100644 index 0000000000000000000000000000000000000000..94f7a95dab9b9ec6ae9e6e81dc51aaa51fd7864b --- /dev/null +++ b/foundational_analysis/cases/case30/case30_23.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.4066396519417586, + "tolerance": 0.020331982597087933 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.46709948828557846, + "tolerance": 0.023354974414278924 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.41113913323246215, + "tolerance": 0.020556956661623108 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.33354391417470547, + "tolerance": 0.016677195708735274 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is theta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_24.json b/foundational_analysis/cases/case30/case30_24.json new file mode 100644 index 0000000000000000000000000000000000000000..cc80c2c3e4ef3dd3419a008d26b692b962cb5526 --- /dev/null +++ b/foundational_analysis/cases/case30/case30_24.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.44731592900399664, + "tolerance": 0.022365796450199833 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.46059879486941946, + "tolerance": 0.023029939743470975 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.4682206394941131, + "tolerance": 0.023411031974705655 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.43154804554426596, + "tolerance": 0.0215774022772133 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is alpha. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case30/case30_25.json b/foundational_analysis/cases/case30/case30_25.json new file mode 100644 index 0000000000000000000000000000000000000000..0a6c1bbf3f0dd146ac4163616ea1314fbd07d772 --- /dev/null +++ b/foundational_analysis/cases/case30/case30_25.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please select standard EEG channels. For each delta, theta, alpha, and beta band, apply a zero-phase FIR bandpass filter to the selected continuous signals, then use the first 120 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform, compute PLV for every unique channel pair, and define the global mean PLV as the arithmetic mean across those pairs. Report the band with the highest global mean PLV and briefly explain the result." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract five outputs from the agent report:\n1) delta global mean PLV\n2) theta global mean PLV\n3) alpha global mean PLV\n4) beta global mean PLV\n5) dominant synchronization band\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"delta_global_mean_plv\", \"theta_global_mean_plv\", \"alpha_global_mean_plv\", \"beta_global_mean_plv\", and \"dominant_synchronization_band\".\n5. The four PLV values must be float or null.\n6. \"dominant_synchronization_band\" must be one of \"delta\", \"theta\", \"alpha\", \"beta\", or null.\n\n### OUTPUT TEMPLATE\n{\"delta_global_mean_plv\": , \"theta_global_mean_plv\": , \"alpha_global_mean_plv\": , \"beta_global_mean_plv\": , \"dominant_synchronization_band\": }", + "metrics": [ + { + "metric_id": "delta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "delta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.43857313163741396, + "tolerance": 0.0219286565818707 + } + }, + { + "metric_id": "theta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "theta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.47502580854732407, + "tolerance": 0.023751290427366206 + } + }, + { + "metric_id": "alpha_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "alpha_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.4225789120158163, + "tolerance": 0.021128945600790816 + } + }, + { + "metric_id": "beta_global_mean_plv_accuracy", + "type": "numeric_check", + "target_key": "beta_global_mean_plv", + "weight": 15, + "params": { + "gt_value": 0.39224183207590985, + "tolerance": 0.019612091603795494 + } + }, + { + "metric_id": "dominant_band_explanation_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "Judge whether the final dominant synchronization band and its explanation are consistent with the true ranking of delta/theta/alpha/beta global mean PLV values. The true dominant synchronization band is theta. Mark the response as correct only if it selects the true highest-PLV band and the explanation clearly supports that band as the strongest synchronization band. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} diff --git a/foundational_analysis/cases/case31/case31_01.json b/foundational_analysis/cases/case31/case31_01.json new file mode 100644 index 0000000000000000000000000000000000000000..07dac0a41617df554543ccf1f6d2118c05a468e0 --- /dev/null +++ b/foundational_analysis/cases/case31/case31_01.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 0.9949164929332257, + "tolerance": 0.04974582464666129 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 1.7065779828281904, + "tolerance": 0.08532889914140952 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention becomes more distracted. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_02.json b/foundational_analysis/cases/case31/case31_02.json new file mode 100644 index 0000000000000000000000000000000000000000..4d737cc97667f61d5fc09913e47456d87d1ece60 --- /dev/null +++ b/foundational_analysis/cases/case31/case31_02.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 0.33353359539813193, + "tolerance": 0.016676679769906597 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 0.5646207863086754, + "tolerance": 0.028231039315433772 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention becomes more distracted. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_03.json b/foundational_analysis/cases/case31/case31_03.json new file mode 100644 index 0000000000000000000000000000000000000000..71089f37b8d825655c13502eae5b13561f962480 --- /dev/null +++ b/foundational_analysis/cases/case31/case31_03.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 0.7090278388543377, + "tolerance": 0.03545139194271689 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 1.5880084351465544, + "tolerance": 0.07940042175732773 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention becomes more distracted. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_04.json b/foundational_analysis/cases/case31/case31_04.json new file mode 100644 index 0000000000000000000000000000000000000000..3f54a7dca0e586a4a79cf461b98396e016fee96b --- /dev/null +++ b/foundational_analysis/cases/case31/case31_04.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 0.2230006149811404, + "tolerance": 0.011150030749057021 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 0.8449826398458985, + "tolerance": 0.04224913199229493 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention becomes more distracted. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_05.json b/foundational_analysis/cases/case31/case31_05.json new file mode 100644 index 0000000000000000000000000000000000000000..f58ae35864a755cc2edfa611a383355fcd1b3a0d --- /dev/null +++ b/foundational_analysis/cases/case31/case31_05.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 0.387586593937157, + "tolerance": 0.01937932969685785 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 0.3430245470691804, + "tolerance": 0.01715122735345902 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention becomes more focused. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_06.json b/foundational_analysis/cases/case31/case31_06.json new file mode 100644 index 0000000000000000000000000000000000000000..d780bd45f41e8606bf53f4daa3138e9ce7dc97ed --- /dev/null +++ b/foundational_analysis/cases/case31/case31_06.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 0.431956493319918, + "tolerance": 0.0215978246659959 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 0.4287180440931179, + "tolerance": 0.021435902204655896 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention remains stable. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_07.json b/foundational_analysis/cases/case31/case31_07.json new file mode 100644 index 0000000000000000000000000000000000000000..180c8ad54368fcfb780ea8a05b41a2e63d183be5 --- /dev/null +++ b/foundational_analysis/cases/case31/case31_07.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 0.20190592092816229, + "tolerance": 0.010095296046408115 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 0.1958422209204513, + "tolerance": 0.009792111046022567 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention remains stable. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_08.json b/foundational_analysis/cases/case31/case31_08.json new file mode 100644 index 0000000000000000000000000000000000000000..7b09b58768dbc898e46f9dcf451dafa5862ef6b4 --- /dev/null +++ b/foundational_analysis/cases/case31/case31_08.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 0.3133082376952488, + "tolerance": 0.01566541188476244 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 0.28129800999209126, + "tolerance": 0.014064900499604564 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention becomes more focused. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_09.json b/foundational_analysis/cases/case31/case31_09.json new file mode 100644 index 0000000000000000000000000000000000000000..396ac572e50ee9952692511b9a706969c28ccfd4 --- /dev/null +++ b/foundational_analysis/cases/case31/case31_09.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 0.4494474614556632, + "tolerance": 0.022472373072783162 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 0.4627682627780333, + "tolerance": 0.023138413138901665 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention remains stable. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_10.json b/foundational_analysis/cases/case31/case31_10.json new file mode 100644 index 0000000000000000000000000000000000000000..48078b61b13435f523be77df3b63207a4df5ef80 --- /dev/null +++ b/foundational_analysis/cases/case31/case31_10.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 0.13216176724798667, + "tolerance": 0.006608088362399334 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 0.15369919721179115, + "tolerance": 0.007684959860589558 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention becomes more distracted. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_11.json b/foundational_analysis/cases/case31/case31_11.json new file mode 100644 index 0000000000000000000000000000000000000000..9aca7c4343495b20966539f1db3030daa1a35920 --- /dev/null +++ b/foundational_analysis/cases/case31/case31_11.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 0.5547909748338024, + "tolerance": 0.02773954874169012 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 0.5764557244504872, + "tolerance": 0.028822786222524363 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention remains stable. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_12.json b/foundational_analysis/cases/case31/case31_12.json new file mode 100644 index 0000000000000000000000000000000000000000..75dbc9eabd7d8e01552486a073a13813136c7b9e --- /dev/null +++ b/foundational_analysis/cases/case31/case31_12.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 0.6397366237928053, + "tolerance": 0.03198683118964026 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 0.6360226335911848, + "tolerance": 0.031801131679559246 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention remains stable. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_13.json b/foundational_analysis/cases/case31/case31_13.json new file mode 100644 index 0000000000000000000000000000000000000000..f75c82f09d897d8a610da6b1e7684c3976b5c559 --- /dev/null +++ b/foundational_analysis/cases/case31/case31_13.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 0.9412574260672343, + "tolerance": 0.04706287130336172 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 0.9057854456944139, + "tolerance": 0.0452892722847207 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention remains stable. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_14.json b/foundational_analysis/cases/case31/case31_14.json new file mode 100644 index 0000000000000000000000000000000000000000..8b605c6f802999361baf980200fe095d14d5336f --- /dev/null +++ b/foundational_analysis/cases/case31/case31_14.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 1.000227215945559, + "tolerance": 0.05001136079727795 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 0.8874530252362344, + "tolerance": 0.044372651261811724 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention becomes more focused. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_15.json b/foundational_analysis/cases/case31/case31_15.json new file mode 100644 index 0000000000000000000000000000000000000000..0011eb2fc6210b71b0d9b1e7f42ac81264092ab1 --- /dev/null +++ b/foundational_analysis/cases/case31/case31_15.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 1.3145838833295465, + "tolerance": 0.06572919416647732 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 1.2576314087050215, + "tolerance": 0.06288157043525108 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention remains stable. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_16.json b/foundational_analysis/cases/case31/case31_16.json new file mode 100644 index 0000000000000000000000000000000000000000..44b2cc13f1f342d018e1f1ecd1332c1566bd24c9 --- /dev/null +++ b/foundational_analysis/cases/case31/case31_16.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 0.8147166873388965, + "tolerance": 0.04073583436694483 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 0.656645635091342, + "tolerance": 0.0328322817545671 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention becomes more focused. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_17.json b/foundational_analysis/cases/case31/case31_17.json new file mode 100644 index 0000000000000000000000000000000000000000..53acba788959904ad9e4194463613d46bf7a401d --- /dev/null +++ b/foundational_analysis/cases/case31/case31_17.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 3.448028716664601, + "tolerance": 0.17240143583323006 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 3.6053980300095865, + "tolerance": 0.18026990150047933 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention remains stable. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_18.json b/foundational_analysis/cases/case31/case31_18.json new file mode 100644 index 0000000000000000000000000000000000000000..f11a8529421299ca75e416f8e59fe0a390cedef2 --- /dev/null +++ b/foundational_analysis/cases/case31/case31_18.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 7.62526242919263, + "tolerance": 0.38126312145963154 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 7.019366069294587, + "tolerance": 0.35096830346472935 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention becomes more focused. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_19.json b/foundational_analysis/cases/case31/case31_19.json new file mode 100644 index 0000000000000000000000000000000000000000..8097fa1ea8d058953f5a0c0eac0bceef4ff803bf --- /dev/null +++ b/foundational_analysis/cases/case31/case31_19.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 3.0153394559554343, + "tolerance": 0.15076697279777174 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 3.0517737100312203, + "tolerance": 0.15258868550156102 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention remains stable. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_20.json b/foundational_analysis/cases/case31/case31_20.json new file mode 100644 index 0000000000000000000000000000000000000000..875da31b2e5f906e75ee341b47449165a093ce02 --- /dev/null +++ b/foundational_analysis/cases/case31/case31_20.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 4.499942748207541, + "tolerance": 0.22499713741037708 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 4.118302939765865, + "tolerance": 0.20591514698829327 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention becomes more focused. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_21.json b/foundational_analysis/cases/case31/case31_21.json new file mode 100644 index 0000000000000000000000000000000000000000..fc8f87624734479ec91e980b815eadd44af1f155 --- /dev/null +++ b/foundational_analysis/cases/case31/case31_21.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 2.595354688553481, + "tolerance": 0.12976773442767406 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 2.703123233041706, + "tolerance": 0.1351561616520853 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention remains stable. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_22.json b/foundational_analysis/cases/case31/case31_22.json new file mode 100644 index 0000000000000000000000000000000000000000..b7dbfc32a7e566e7e6e6c8f628db99a9e86e9bd3 --- /dev/null +++ b/foundational_analysis/cases/case31/case31_22.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 1.0437813350889409, + "tolerance": 0.052189066754447046 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 1.2573438098179577, + "tolerance": 0.06286719049089788 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention becomes more distracted. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_23.json b/foundational_analysis/cases/case31/case31_23.json new file mode 100644 index 0000000000000000000000000000000000000000..2e5623f2263b34c8a18165c18a6fa8677aa9a508 --- /dev/null +++ b/foundational_analysis/cases/case31/case31_23.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 0.9234181535771928, + "tolerance": 0.046170907678859646 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 1.089100446761241, + "tolerance": 0.054455022338062056 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention becomes more distracted. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_24.json b/foundational_analysis/cases/case31/case31_24.json new file mode 100644 index 0000000000000000000000000000000000000000..ac82b28d6bbc140f52417d894c3d90fff478e121 --- /dev/null +++ b/foundational_analysis/cases/case31/case31_24.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 3.389427347572558, + "tolerance": 0.1694713673786279 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 3.055852290180714, + "tolerance": 0.1527926145090357 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention becomes more focused. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case31/case31_25.json b/foundational_analysis/cases/case31/case31_25.json new file mode 100644 index 0000000000000000000000000000000000000000..8885f8a50ad241a7c775b92d64bca41c88f9f5a8 --- /dev/null +++ b/foundational_analysis/cases/case31/case31_25.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please select frontal-region channels from the standard EEG channels, use the first 60 seconds of data, apply a 0.5-40Hz FIR bandpass filter, and split the filtered data into two 30-second windows. For each window, use Welch PSD estimation and integrate the PSD over non-overlapping alpha [8, 13) Hz and beta [13, 30] Hz bands. Sum each band's power across the selected channels, then compute ABR as summed alpha power divided by summed beta power. Report both ABR values and explain whether the attention state becomes more focused, more distracted, or stable. Define stable as the window-2 ABR being within 5% relative change of the window-1 ABR." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two ABR values from the agent report:\n1) ABR for the first 30-second window\n2) ABR for the second 30-second window\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"abr_window_1\" and \"abr_window_2\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"abr_window_1\": , \"abr_window_2\": }", + "metrics": [ + { + "metric_id": "abr_window_1_accuracy", + "type": "numeric_check", + "target_key": "abr_window_1", + "weight": 30, + "params": { + "gt_value": 0.7469967835814567, + "tolerance": 0.037349839179072836 + } + }, + { + "metric_id": "abr_window_2_accuracy", + "type": "numeric_check", + "target_key": "abr_window_2", + "weight": 30, + "params": { + "gt_value": 1.2695403045306792, + "tolerance": 0.06347701522653397 + } + }, + { + "metric_id": "attention_state_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG attention-state task based on frontal ABR changes. The true attention-state change is: attention becomes more distracted. Use the stated true attention-state change as the ground truth. Mark the response as correct only if its explanation matches this true attention-state change. Mark it as incorrect if it claims the opposite trend, claims stability when the true state changes, or claims a change when the true state is stable. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_01.json b/foundational_analysis/cases/case32/case32_01.json new file mode 100644 index 0000000000000000000000000000000000000000..4de970ba76f456fd8d5dfacd0b0ccb7f67abb210 --- /dev/null +++ b/foundational_analysis/cases/case32/case32_01.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 3 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 3 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_02.json b/foundational_analysis/cases/case32/case32_02.json new file mode 100644 index 0000000000000000000000000000000000000000..bcf4f449a8d16441a2da7b7337914a8fd68f6f7f --- /dev/null +++ b/foundational_analysis/cases/case32/case32_02.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 4 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 3 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric masks regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_03.json b/foundational_analysis/cases/case32/case32_03.json new file mode 100644 index 0000000000000000000000000000000000000000..f7638c537123521dfbef9ab0c7dd3ed271281c4b --- /dev/null +++ b/foundational_analysis/cases/case32/case32_03.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 3 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 3 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_04.json b/foundational_analysis/cases/case32/case32_04.json new file mode 100644 index 0000000000000000000000000000000000000000..45fcad2977ef6fbc48d803de5cfc653755b30698 --- /dev/null +++ b/foundational_analysis/cases/case32/case32_04.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 4 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 4 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_05.json b/foundational_analysis/cases/case32/case32_05.json new file mode 100644 index 0000000000000000000000000000000000000000..20e2f22e95b8556f5b2f6afb7cb44ce3e963f67d --- /dev/null +++ b/foundational_analysis/cases/case32/case32_05.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 4 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 4 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_06.json b/foundational_analysis/cases/case32/case32_06.json new file mode 100644 index 0000000000000000000000000000000000000000..c00d743f4a5d5db32e7436c77b1623de67f2460e --- /dev/null +++ b/foundational_analysis/cases/case32/case32_06.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 3 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 3 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_07.json b/foundational_analysis/cases/case32/case32_07.json new file mode 100644 index 0000000000000000000000000000000000000000..e41a9475d012d9bdabb5ace94078eb182c9e555d --- /dev/null +++ b/foundational_analysis/cases/case32/case32_07.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 3 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 4 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric masks regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_08.json b/foundational_analysis/cases/case32/case32_08.json new file mode 100644 index 0000000000000000000000000000000000000000..73465aec1d0819e02097467ffc4e56b8c5a49d95 --- /dev/null +++ b/foundational_analysis/cases/case32/case32_08.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 4 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 4 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_09.json b/foundational_analysis/cases/case32/case32_09.json new file mode 100644 index 0000000000000000000000000000000000000000..2fec992690b0f62b0d1de3204af9faa00dd9bcd8 --- /dev/null +++ b/foundational_analysis/cases/case32/case32_09.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 3 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 3 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_10.json b/foundational_analysis/cases/case32/case32_10.json new file mode 100644 index 0000000000000000000000000000000000000000..7860665740df610f537a222cc4f28fffa8ad8ac3 --- /dev/null +++ b/foundational_analysis/cases/case32/case32_10.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 3 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 3 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_11.json b/foundational_analysis/cases/case32/case32_11.json new file mode 100644 index 0000000000000000000000000000000000000000..d1c3075d21b5ffcdb18b97d111f7d36ab0f1e3f9 --- /dev/null +++ b/foundational_analysis/cases/case32/case32_11.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 5 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 5 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_12.json b/foundational_analysis/cases/case32/case32_12.json new file mode 100644 index 0000000000000000000000000000000000000000..46333415d7db4e436cd23406f14686358c3f1bb3 --- /dev/null +++ b/foundational_analysis/cases/case32/case32_12.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 4 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 4 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_13.json b/foundational_analysis/cases/case32/case32_13.json new file mode 100644 index 0000000000000000000000000000000000000000..bbe090f29d3abedd27d71ac3335493368fc38131 --- /dev/null +++ b/foundational_analysis/cases/case32/case32_13.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 4 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 4 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_14.json b/foundational_analysis/cases/case32/case32_14.json new file mode 100644 index 0000000000000000000000000000000000000000..cadbcdc4edb1f8a98bcbf14a328d0bf217c79ea3 --- /dev/null +++ b/foundational_analysis/cases/case32/case32_14.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 3 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 3 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_15.json b/foundational_analysis/cases/case32/case32_15.json new file mode 100644 index 0000000000000000000000000000000000000000..1a8e3282081e06867233c56287bf10b500c01501 --- /dev/null +++ b/foundational_analysis/cases/case32/case32_15.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 3 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 3 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_16.json b/foundational_analysis/cases/case32/case32_16.json new file mode 100644 index 0000000000000000000000000000000000000000..bea7bef867eedf666ddfa263fc47c203dbad63c7 --- /dev/null +++ b/foundational_analysis/cases/case32/case32_16.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 2 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 1 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric masks regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_17.json b/foundational_analysis/cases/case32/case32_17.json new file mode 100644 index 0000000000000000000000000000000000000000..64ea1ee889be39665da5fab52c8054f4a21bd1c8 --- /dev/null +++ b/foundational_analysis/cases/case32/case32_17.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 2 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 1 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric masks regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_18.json b/foundational_analysis/cases/case32/case32_18.json new file mode 100644 index 0000000000000000000000000000000000000000..ab92579b961899cbd0fc22041f5939fd5b4cb225 --- /dev/null +++ b/foundational_analysis/cases/case32/case32_18.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 1 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 1 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_19.json b/foundational_analysis/cases/case32/case32_19.json new file mode 100644 index 0000000000000000000000000000000000000000..a8b14b2f68255558e947b811b2a73b347f5ac141 --- /dev/null +++ b/foundational_analysis/cases/case32/case32_19.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 2 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 1 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric masks regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_20.json b/foundational_analysis/cases/case32/case32_20.json new file mode 100644 index 0000000000000000000000000000000000000000..bd1fc67ceceec1919f08053787ca004702fff74e --- /dev/null +++ b/foundational_analysis/cases/case32/case32_20.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 1 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 1 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_21.json b/foundational_analysis/cases/case32/case32_21.json new file mode 100644 index 0000000000000000000000000000000000000000..767e6687f6e987765f172203446f430173c988ba --- /dev/null +++ b/foundational_analysis/cases/case32/case32_21.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 1 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 1 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_22.json b/foundational_analysis/cases/case32/case32_22.json new file mode 100644 index 0000000000000000000000000000000000000000..481d711e7a9f5a14f49d9a5b979ba0ab157c027a --- /dev/null +++ b/foundational_analysis/cases/case32/case32_22.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 2 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 2 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_23.json b/foundational_analysis/cases/case32/case32_23.json new file mode 100644 index 0000000000000000000000000000000000000000..a9038a50de3785f5aaff17f6a63cbeb3d9ae3bc7 --- /dev/null +++ b/foundational_analysis/cases/case32/case32_23.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 2 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 2 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_24.json b/foundational_analysis/cases/case32/case32_24.json new file mode 100644 index 0000000000000000000000000000000000000000..6022391e4853842b2fca6906282b03c6a7f779c2 --- /dev/null +++ b/foundational_analysis/cases/case32/case32_24.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 1 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 1 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case32/case32_25.json b/foundational_analysis/cases/case32/case32_25.json new file mode 100644 index 0000000000000000000000000000000000000000..df117104078c2c475f71b5662e75a378e53df339 --- /dev/null +++ b/foundational_analysis/cases/case32/case32_25.json @@ -0,0 +1,44 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please select standard EEG channels and rank alpha relative power among the delta, theta, alpha, beta, and gamma bands. Compute the global rank from the mean relative power across all standard EEG channels, then compute the occipital rank from the mean relative power across occipital channels only. Each rank must be an integer from 1 to 5, where 1 is the highest and 5 is the lowest. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz. Determine whether the global average metric masks regional differences, and briefly explain your conclusion." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two alpha relative power ranks from the agent report:\n1) global alpha relative power rank\n2) occipital alpha relative power rank\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"global_alpha_rank\" and \"occipital_alpha_rank\".\n5. Each value must be an integer from 1 to 5, or null if missing.\n\n### OUTPUT TEMPLATE\n{\"global_alpha_rank\": , \"occipital_alpha_rank\": }", + "metrics": [ + { + "metric_id": "global_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "global_alpha_rank", + "weight": 30, + "params": { + "gt_value": 2 + } + }, + { + "metric_id": "occipital_alpha_rank_accuracy", + "type": "categorical_check", + "target_key": "occipital_alpha_rank", + "weight": 30, + "params": { + "gt_value": 2 + } + }, + { + "metric_id": "global_occipital_masking_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG task comparing global and occipital alpha relative power ranks. The true conclusion is: the global average metric does not mask regional indicator differences. Use the stated true conclusion as the ground truth; do not infer a different conclusion. Mark the response as correct only if its explanation matches this true conclusion. Mark it as incorrect if it claims masking when the true conclusion says no masking, or claims no masking when the true conclusion says masking. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_01.json b/foundational_analysis/cases/case33/case33_01.json new file mode 100644 index 0000000000000000000000000000000000000000..c89cd6a14f87edc83cda22f7e9eb8c787e85c7c1 --- /dev/null +++ b/foundational_analysis/cases/case33/case33_01.json @@ -0,0 +1,32 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "O1", + "O2" + ], + "match_mode": "per_hit_fraction", + "top_k": 2, + "score_per_hit": 0.5 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_02.json b/foundational_analysis/cases/case33/case33_02.json new file mode 100644 index 0000000000000000000000000000000000000000..144bac22499e9b12bd9c31508a080bb82ee2b2a5 --- /dev/null +++ b/foundational_analysis/cases/case33/case33_02.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "C4", + "F3", + "F4" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_03.json b/foundational_analysis/cases/case33/case33_03.json new file mode 100644 index 0000000000000000000000000000000000000000..05547655ff7d20638a64f06c988c654df5fbbe24 --- /dev/null +++ b/foundational_analysis/cases/case33/case33_03.json @@ -0,0 +1,32 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "O1", + "O2" + ], + "match_mode": "per_hit_fraction", + "top_k": 2, + "score_per_hit": 0.5 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_04.json b/foundational_analysis/cases/case33/case33_04.json new file mode 100644 index 0000000000000000000000000000000000000000..68cb1629902551d6dd5d9fb64c22a39963578030 --- /dev/null +++ b/foundational_analysis/cases/case33/case33_04.json @@ -0,0 +1,32 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "O1", + "O2" + ], + "match_mode": "per_hit_fraction", + "top_k": 2, + "score_per_hit": 0.5 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_05.json b/foundational_analysis/cases/case33/case33_05.json new file mode 100644 index 0000000000000000000000000000000000000000..122397f72c9bc4c1d0f3cc969bd6d24296594e65 --- /dev/null +++ b/foundational_analysis/cases/case33/case33_05.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "C3", + "C4", + "F3" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_06.json b/foundational_analysis/cases/case33/case33_06.json new file mode 100644 index 0000000000000000000000000000000000000000..a23ec615c2d0f17fde5e1ab56c40ff7267b088fc --- /dev/null +++ b/foundational_analysis/cases/case33/case33_06.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "C1", + "C2", + "C3", + "C4", + "C5", + "C6", + "CP2", + "CZ", + "FC1", + "FC2", + "FC3", + "FC5", + "FT7", + "FZ", + "O1", + "OZ", + "P2", + "P3", + "P4", + "P6", + "P7", + "PO10", + "PO3", + "PO4", + "PO7", + "PO8", + "PZ", + "T7", + "TP7" + ], + "match_mode": "per_hit_fraction", + "top_k": 29, + "score_per_hit": 0.034482758620689655 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_07.json b/foundational_analysis/cases/case33/case33_07.json new file mode 100644 index 0000000000000000000000000000000000000000..63ca5d36d0a09a446b9d7f4569c629fa5f690182 --- /dev/null +++ b/foundational_analysis/cases/case33/case33_07.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "AF3", + "AF4", + "C2", + "C3", + "C4", + "C6", + "CP1", + "F1", + "F3", + "F4", + "FC3", + "FC5", + "FZ", + "O1", + "P2", + "P3", + "P4", + "P6", + "P7", + "P8", + "PO3", + "PO9", + "POZ", + "T7", + "T8", + "TP10", + "TP7", + "TP8", + "TP9" + ], + "match_mode": "per_hit_fraction", + "top_k": 29, + "score_per_hit": 0.034482758620689655 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_08.json b/foundational_analysis/cases/case33/case33_08.json new file mode 100644 index 0000000000000000000000000000000000000000..36ca15f3b75848dbd8f2dcdc795bb8df6c629fa3 --- /dev/null +++ b/foundational_analysis/cases/case33/case33_08.json @@ -0,0 +1,54 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "C1", + "C3", + "C5", + "C6", + "CP6", + "CPZ", + "F1", + "F3", + "FC2", + "FC5", + "FC6", + "FT8", + "O1", + "O2", + "OZ", + "P2", + "P5", + "P7", + "P8", + "PO10", + "PO3", + "PO7", + "PO8", + "T8" + ], + "match_mode": "per_hit_fraction", + "top_k": 24, + "score_per_hit": 0.041666666666666664 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_09.json b/foundational_analysis/cases/case33/case33_09.json new file mode 100644 index 0000000000000000000000000000000000000000..c74b36717fd787f7017966b28fbf15ee1fff52aa --- /dev/null +++ b/foundational_analysis/cases/case33/case33_09.json @@ -0,0 +1,62 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "AF3", + "AF4", + "AF8", + "C1", + "C2", + "C3", + "C4", + "C5", + "C6", + "CP1", + "CP2", + "CP3", + "CP6", + "CPZ", + "CZ", + "F1", + "F2", + "F3", + "F5", + "FC3", + "FC4", + "FC5", + "FC6", + "FP1", + "FP2", + "FZ", + "P2", + "PO9", + "PZ", + "T8", + "TP7", + "TP8" + ], + "match_mode": "per_hit_fraction", + "top_k": 32, + "score_per_hit": 0.03125 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_10.json b/foundational_analysis/cases/case33/case33_10.json new file mode 100644 index 0000000000000000000000000000000000000000..8c66e0444ac39e868659ab482f764338529c5a12 --- /dev/null +++ b/foundational_analysis/cases/case33/case33_10.json @@ -0,0 +1,56 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "AF8", + "C1", + "C3", + "C4", + "C5", + "CP2", + "CP4", + "CP6", + "F3", + "F6", + "F8", + "FC3", + "FC4", + "FT10", + "P1", + "P3", + "P4", + "P5", + "PO10", + "PO3", + "PO4", + "PO9", + "POZ", + "T7", + "T8", + "TP8" + ], + "match_mode": "per_hit_fraction", + "top_k": 26, + "score_per_hit": 0.038461538461538464 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_11.json b/foundational_analysis/cases/case33/case33_11.json new file mode 100644 index 0000000000000000000000000000000000000000..5a60afb0caec1e433ff18463af2da36892b5bd0f --- /dev/null +++ b/foundational_analysis/cases/case33/case33_11.json @@ -0,0 +1,59 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "C1", + "C2", + "C3", + "C4", + "CP1", + "CP2", + "CP3", + "CP4", + "CP5", + "CP6", + "CPZ", + "CZ", + "O1", + "O2", + "OZ", + "P1", + "P2", + "P3", + "P4", + "P5", + "P6", + "P7", + "P8", + "PO3", + "PO4", + "PO5", + "POZ", + "PZ", + "TP7" + ], + "match_mode": "per_hit_fraction", + "top_k": 29, + "score_per_hit": 0.034482758620689655 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_12.json b/foundational_analysis/cases/case33/case33_12.json new file mode 100644 index 0000000000000000000000000000000000000000..ba868114966b392fd70edd60e1a322339e492aab --- /dev/null +++ b/foundational_analysis/cases/case33/case33_12.json @@ -0,0 +1,57 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "C1", + "C2", + "C3", + "C4", + "C5", + "CP1", + "CP2", + "CP3", + "CP4", + "CP5", + "CPZ", + "CZ", + "FC1", + "FC2", + "FC4", + "FCZ", + "FZ", + "P1", + "P2", + "P3", + "P6", + "PO3", + "PO5", + "PO6", + "PO7", + "POZ", + "PZ" + ], + "match_mode": "per_hit_fraction", + "top_k": 27, + "score_per_hit": 0.037037037037037035 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_13.json b/foundational_analysis/cases/case33/case33_13.json new file mode 100644 index 0000000000000000000000000000000000000000..a68071d882f013f3ebf4254f4f9fec1e722a83d6 --- /dev/null +++ b/foundational_analysis/cases/case33/case33_13.json @@ -0,0 +1,61 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "C3", + "C4", + "C6", + "CB2", + "CP1", + "CP2", + "CP3", + "CP4", + "CP6", + "CPZ", + "FC2", + "FC4", + "O1", + "O2", + "OZ", + "P1", + "P2", + "P3", + "P4", + "P5", + "P6", + "P7", + "P8", + "PO3", + "PO4", + "PO5", + "PO6", + "PO7", + "PO8", + "POZ", + "PZ" + ], + "match_mode": "per_hit_fraction", + "top_k": 31, + "score_per_hit": 0.03225806451612903 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_14.json b/foundational_analysis/cases/case33/case33_14.json new file mode 100644 index 0000000000000000000000000000000000000000..470740a72b3e1b26e5554837542543564fc95c7f --- /dev/null +++ b/foundational_analysis/cases/case33/case33_14.json @@ -0,0 +1,66 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "C2", + "C3", + "C4", + "C5", + "CP1", + "CP2", + "CP3", + "CP4", + "CP5", + "CPZ", + "F1", + "FC1", + "FC2", + "FC3", + "FCZ", + "FZ", + "O1", + "O2", + "OZ", + "P1", + "P2", + "P3", + "P4", + "P5", + "P6", + "P7", + "P8", + "PO3", + "PO4", + "PO5", + "PO6", + "PO7", + "PO8", + "POZ", + "PZ", + "TP7" + ], + "match_mode": "per_hit_fraction", + "top_k": 36, + "score_per_hit": 0.027777777777777776 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_15.json b/foundational_analysis/cases/case33/case33_15.json new file mode 100644 index 0000000000000000000000000000000000000000..36e1544ca503ff63bf151a906e56722f97c1a06d --- /dev/null +++ b/foundational_analysis/cases/case33/case33_15.json @@ -0,0 +1,60 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "C1", + "C2", + "C3", + "C4", + "CP1", + "CP2", + "CP3", + "CP4", + "CP5", + "CP6", + "O1", + "O2", + "OZ", + "P1", + "P2", + "P3", + "P4", + "P5", + "P6", + "P7", + "P8", + "PO3", + "PO4", + "PO5", + "PO6", + "PO7", + "PO8", + "POZ", + "PZ", + "TP7" + ], + "match_mode": "per_hit_fraction", + "top_k": 30, + "score_per_hit": 0.03333333333333333 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_16.json b/foundational_analysis/cases/case33/case33_16.json new file mode 100644 index 0000000000000000000000000000000000000000..c00dcb685d0ebef11ad972694b489a10c0592bd5 --- /dev/null +++ b/foundational_analysis/cases/case33/case33_16.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "C3", + "O1", + "O2", + "P3", + "P4", + "P7", + "P8", + "PZ" + ], + "match_mode": "per_hit_fraction", + "top_k": 8, + "score_per_hit": 0.125 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_17.json b/foundational_analysis/cases/case33/case33_17.json new file mode 100644 index 0000000000000000000000000000000000000000..eb7355e794a2897249f4c040af6a06ca4089a9f0 --- /dev/null +++ b/foundational_analysis/cases/case33/case33_17.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "C4", + "CZ", + "O1", + "O2", + "P3", + "P4", + "P7", + "PZ" + ], + "match_mode": "per_hit_fraction", + "top_k": 8, + "score_per_hit": 0.125 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_18.json b/foundational_analysis/cases/case33/case33_18.json new file mode 100644 index 0000000000000000000000000000000000000000..7d2a7a69c710f55d5c36299f55e71a6dff5d9f30 --- /dev/null +++ b/foundational_analysis/cases/case33/case33_18.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "C3", + "O1", + "O2", + "P3", + "P4", + "P7", + "P8", + "PZ" + ], + "match_mode": "per_hit_fraction", + "top_k": 8, + "score_per_hit": 0.125 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_19.json b/foundational_analysis/cases/case33/case33_19.json new file mode 100644 index 0000000000000000000000000000000000000000..ff7d820e8f8d686ffb9787df6daa642c3bffd630 --- /dev/null +++ b/foundational_analysis/cases/case33/case33_19.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "C4", + "CZ", + "O1", + "O2", + "P3", + "P7" + ], + "match_mode": "per_hit_fraction", + "top_k": 6, + "score_per_hit": 0.16666666666666666 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_20.json b/foundational_analysis/cases/case33/case33_20.json new file mode 100644 index 0000000000000000000000000000000000000000..d8f9b2e2c06abe80c6de37bb4de6401e511d606e --- /dev/null +++ b/foundational_analysis/cases/case33/case33_20.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "O1", + "O2", + "P3", + "P4", + "P7", + "P8", + "PZ" + ], + "match_mode": "per_hit_fraction", + "top_k": 7, + "score_per_hit": 0.14285714285714285 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_21.json b/foundational_analysis/cases/case33/case33_21.json new file mode 100644 index 0000000000000000000000000000000000000000..39651f45ef61a2075186ab14363b172dca96feb9 --- /dev/null +++ b/foundational_analysis/cases/case33/case33_21.json @@ -0,0 +1,40 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "F3", + "F4", + "FP1", + "FZ", + "O1", + "P3", + "P4", + "P8", + "PZ", + "T8" + ], + "match_mode": "per_hit_fraction", + "top_k": 10, + "score_per_hit": 0.1 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_22.json b/foundational_analysis/cases/case33/case33_22.json new file mode 100644 index 0000000000000000000000000000000000000000..55ac3512a7ea1be7fe3bfea395256419851f3fd6 --- /dev/null +++ b/foundational_analysis/cases/case33/case33_22.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "F7", + "FP1", + "FP2", + "FZ", + "O2", + "P3", + "P4" + ], + "match_mode": "per_hit_fraction", + "top_k": 7, + "score_per_hit": 0.14285714285714285 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_23.json b/foundational_analysis/cases/case33/case33_23.json new file mode 100644 index 0000000000000000000000000000000000000000..61aef026d33d1a2c76b032731c914e2688d967e9 --- /dev/null +++ b/foundational_analysis/cases/case33/case33_23.json @@ -0,0 +1,36 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "F3", + "P3", + "P4", + "P7", + "PZ", + "T7" + ], + "match_mode": "per_hit_fraction", + "top_k": 6, + "score_per_hit": 0.16666666666666666 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_24.json b/foundational_analysis/cases/case33/case33_24.json new file mode 100644 index 0000000000000000000000000000000000000000..1ac7e3b864fd69d2d8e0b28df5c5ff487c60e01a --- /dev/null +++ b/foundational_analysis/cases/case33/case33_24.json @@ -0,0 +1,38 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "C4", + "O1", + "O2", + "P3", + "P4", + "P7", + "P8", + "PZ" + ], + "match_mode": "per_hit_fraction", + "top_k": 8, + "score_per_hit": 0.125 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case33/case33_25.json b/foundational_analysis/cases/case33/case33_25.json new file mode 100644 index 0000000000000000000000000000000000000000..140ec16a38391eef04adfcd15b176b975ef8f4f8 --- /dev/null +++ b/foundational_analysis/cases/case33/case33_25.json @@ -0,0 +1,35 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please select standard EEG channels. For each channel, compute the alpha-band relative power and signal variance. Exclude channels with variance below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR. Among the remaining channels, return the set of channels whose alpha relative power is greater than the global mean across the remaining channels. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of qualified EEG channels from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"qualified_channels\".\n5. Extract only the final channels that remain after variance exclusion AND have alpha relative power above the global mean of the remaining channels. The value must be an array of strings.\n6. If the report contains excluded_channels, valid_channels, all_channels, and channels_above_mean (or equivalent labels), use only channels_above_mean. Never output excluded channels or the full valid-channel list.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Examples: EEG C3-A2 -> C3; EEG Fp1-LE -> FP1; EEG T5 -> P7; Pz -> PZ.\n11. Preserve the final set order as reported; do not add channels from intermediate lists.\n12. If the report explicitly says no channel meets both criteria, return an empty array. If the final qualified set is missing or unclear, return {\"qualified_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"qualified_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "qualified_alpha_channel_set", + "type": "set_match_check", + "target_key": "qualified_channels", + "weight": 100, + "params": { + "gt_value": [ + "C3", + "O1", + "P3", + "P4", + "PZ" + ], + "match_mode": "per_hit_fraction", + "top_k": 5, + "score_per_hit": 0.2 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_01.json b/foundational_analysis/cases/case34/case34_01.json new file mode 100644 index 0000000000000000000000000000000000000000..6fb83355aa93a14e4958f3d675d6c91f2638556a --- /dev/null +++ b/foundational_analysis/cases/case34/case34_01.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 8.75, + "tolerance": 0.4375 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 18.337664727390138, + "tolerance": 0.916883236369507 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_02.json b/foundational_analysis/cases/case34/case34_02.json new file mode 100644 index 0000000000000000000000000000000000000000..da4eefe11bfb24018c33c1131deacc9b24c23646 --- /dev/null +++ b/foundational_analysis/cases/case34/case34_02.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 8.0, + "tolerance": 0.4 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 9.559376440534951, + "tolerance": 0.47796882202674756 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_03.json b/foundational_analysis/cases/case34/case34_03.json new file mode 100644 index 0000000000000000000000000000000000000000..239715ac29c1028dd13e3324ce75bc8b0715a52c --- /dev/null +++ b/foundational_analysis/cases/case34/case34_03.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 10.0, + "tolerance": 0.5 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 13.372305601271986, + "tolerance": 0.6686152800635994 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_04.json b/foundational_analysis/cases/case34/case34_04.json new file mode 100644 index 0000000000000000000000000000000000000000..afb1610e690845d22804b4f3405688aad972f6f8 --- /dev/null +++ b/foundational_analysis/cases/case34/case34_04.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 8.0, + "tolerance": 0.4 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 7.453772676514644, + "tolerance": 0.37268863382573225 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_05.json b/foundational_analysis/cases/case34/case34_05.json new file mode 100644 index 0000000000000000000000000000000000000000..9e8d02ceabfd623c6867d83a397d0c54422ce3a0 --- /dev/null +++ b/foundational_analysis/cases/case34/case34_05.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 11.125, + "tolerance": 0.55625 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 6.027191226935336, + "tolerance": 0.3013595613467668 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_06.json b/foundational_analysis/cases/case34/case34_06.json new file mode 100644 index 0000000000000000000000000000000000000000..b34f0b68d076abe048db9abaed2511ae9692feab --- /dev/null +++ b/foundational_analysis/cases/case34/case34_06.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 12.125, + "tolerance": 0.6062500000000001 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 11.521615815731993, + "tolerance": 0.5760807907865997 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_07.json b/foundational_analysis/cases/case34/case34_07.json new file mode 100644 index 0000000000000000000000000000000000000000..4952828b0b0ec8efc172d89cc473138f9609a0d0 --- /dev/null +++ b/foundational_analysis/cases/case34/case34_07.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 10.125, + "tolerance": 0.50625 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 8.283103169535186, + "tolerance": 0.4141551584767593 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_08.json b/foundational_analysis/cases/case34/case34_08.json new file mode 100644 index 0000000000000000000000000000000000000000..0e4bb3825cd8e5a489842c37cb9e0a4e56bcf3af --- /dev/null +++ b/foundational_analysis/cases/case34/case34_08.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 10.05, + "tolerance": 0.5025000000000001 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 9.151526272462233, + "tolerance": 0.4575763136231117 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_09.json b/foundational_analysis/cases/case34/case34_09.json new file mode 100644 index 0000000000000000000000000000000000000000..5fa6526d602e6f8830078acdf025abb880179872 --- /dev/null +++ b/foundational_analysis/cases/case34/case34_09.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 10.85, + "tolerance": 0.5425 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 9.56756263367133, + "tolerance": 0.47837813168356647 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_10.json b/foundational_analysis/cases/case34/case34_10.json new file mode 100644 index 0000000000000000000000000000000000000000..12d17338740d7824c88725bfa593dc2a28e729d8 --- /dev/null +++ b/foundational_analysis/cases/case34/case34_10.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 10.475, + "tolerance": 0.52375 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 9.22722944241766, + "tolerance": 0.46136147212088296 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_11.json b/foundational_analysis/cases/case34/case34_11.json new file mode 100644 index 0000000000000000000000000000000000000000..ff8bb3febc2900b4015a7e18c198d951ae8a9410 --- /dev/null +++ b/foundational_analysis/cases/case34/case34_11.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 8.0, + "tolerance": 0.4 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 5.9144180970303815, + "tolerance": 0.2957209048515191 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_12.json b/foundational_analysis/cases/case34/case34_12.json new file mode 100644 index 0000000000000000000000000000000000000000..1abd23c8cf5ebc985add1c7459fe7ff0f4d56e27 --- /dev/null +++ b/foundational_analysis/cases/case34/case34_12.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 9.4, + "tolerance": 0.47000000000000003 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 4.610367259503859, + "tolerance": 0.230518362975193 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_13.json b/foundational_analysis/cases/case34/case34_13.json new file mode 100644 index 0000000000000000000000000000000000000000..370cdb3e3175a2cd8ff94f7a59c23bcde5b30b4b --- /dev/null +++ b/foundational_analysis/cases/case34/case34_13.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 8.0, + "tolerance": 0.4 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 9.17600005480029, + "tolerance": 0.45880000274001453 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_14.json b/foundational_analysis/cases/case34/case34_14.json new file mode 100644 index 0000000000000000000000000000000000000000..375758077ca9ba35a54b6581452e98d47d7a379c --- /dev/null +++ b/foundational_analysis/cases/case34/case34_14.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 9.25, + "tolerance": 0.4625 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 12.786899875010743, + "tolerance": 0.6393449937505372 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_15.json b/foundational_analysis/cases/case34/case34_15.json new file mode 100644 index 0000000000000000000000000000000000000000..906a1de268c48e414c84c5616b8644b243e49652 --- /dev/null +++ b/foundational_analysis/cases/case34/case34_15.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 10.7, + "tolerance": 0.535 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 9.932797240942474, + "tolerance": 0.49663986204712374 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_16.json b/foundational_analysis/cases/case34/case34_16.json new file mode 100644 index 0000000000000000000000000000000000000000..65a4ca7775e54e5aceaa79eee955f2501c702404 --- /dev/null +++ b/foundational_analysis/cases/case34/case34_16.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 10.25, + "tolerance": 0.5125000000000001 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 37.56568517066687, + "tolerance": 1.8782842585333435 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_17.json b/foundational_analysis/cases/case34/case34_17.json new file mode 100644 index 0000000000000000000000000000000000000000..1f3164fc828af801b5c0a9574425c2ab41ee7652 --- /dev/null +++ b/foundational_analysis/cases/case34/case34_17.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 10.125, + "tolerance": 0.50625 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 58.12224275287765, + "tolerance": 2.9061121376438828 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_18.json b/foundational_analysis/cases/case34/case34_18.json new file mode 100644 index 0000000000000000000000000000000000000000..afe65e0e15bbc27017a18422982128f35746a60d --- /dev/null +++ b/foundational_analysis/cases/case34/case34_18.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 10.375, + "tolerance": 0.51875 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 67.33645419872178, + "tolerance": 3.3668227099360895 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_19.json b/foundational_analysis/cases/case34/case34_19.json new file mode 100644 index 0000000000000000000000000000000000000000..89ac1bc75002b502a9481db545e7198e04b73473 --- /dev/null +++ b/foundational_analysis/cases/case34/case34_19.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 11.0, + "tolerance": 0.55 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 39.25053460368444, + "tolerance": 1.9625267301842222 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_20.json b/foundational_analysis/cases/case34/case34_20.json new file mode 100644 index 0000000000000000000000000000000000000000..c23feae92d005d72bb75dd005014c181729f46bb --- /dev/null +++ b/foundational_analysis/cases/case34/case34_20.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 9.75, + "tolerance": 0.48750000000000004 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 49.618842609196776, + "tolerance": 2.480942130459839 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_21.json b/foundational_analysis/cases/case34/case34_21.json new file mode 100644 index 0000000000000000000000000000000000000000..f747ec97d37ba118009a73b4dbff6be86e038755 --- /dev/null +++ b/foundational_analysis/cases/case34/case34_21.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 10.5, + "tolerance": 0.525 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 55.646380542755495, + "tolerance": 2.782319027137775 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_22.json b/foundational_analysis/cases/case34/case34_22.json new file mode 100644 index 0000000000000000000000000000000000000000..51d451f493f7a1284cc02f16cfa947fac52b83a0 --- /dev/null +++ b/foundational_analysis/cases/case34/case34_22.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 11.0, + "tolerance": 0.55 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 25.968889243653383, + "tolerance": 1.2984444621826692 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_23.json b/foundational_analysis/cases/case34/case34_23.json new file mode 100644 index 0000000000000000000000000000000000000000..b9966f67ff6c02c0fd70b6b62c13e0bdfca9c380 --- /dev/null +++ b/foundational_analysis/cases/case34/case34_23.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 10.875, + "tolerance": 0.5437500000000001 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 30.544567991739846, + "tolerance": 1.5272283995869924 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_24.json b/foundational_analysis/cases/case34/case34_24.json new file mode 100644 index 0000000000000000000000000000000000000000..cd28817e4e17651e9f03c40d2aa84e2e7512f07f --- /dev/null +++ b/foundational_analysis/cases/case34/case34_24.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 10.0, + "tolerance": 0.5 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 56.65364293262889, + "tolerance": 2.832682146631445 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case34/case34_25.json b/foundational_analysis/cases/case34/case34_25.json new file mode 100644 index 0000000000000000000000000000000000000000..db3ea1533279d481d3ac74cd4c877d4128eaf39e --- /dev/null +++ b/foundational_analysis/cases/case34/case34_25.json @@ -0,0 +1,37 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please select occipital EEG channels, compute the mean Alpha Peak Frequency (APF), define the personalized alpha band as APF-2Hz to APF+2Hz, and compute the relative power percentage of this personalized alpha band. Clearly report both the mean APF value and the personalized alpha relative power percentage. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) mean occipital Alpha Peak Frequency (APF)\n2) personalized alpha relative power percentage\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"occipital_apf_mean\" and \"personalized_alpha_ratio\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"occipital_apf_mean\": , \"personalized_alpha_ratio\": }", + "metrics": [ + { + "metric_id": "occipital_apf_mean_accuracy", + "type": "numeric_check", + "target_key": "occipital_apf_mean", + "weight": 50, + "params": { + "gt_value": 9.75, + "tolerance": 0.48750000000000004 + } + }, + { + "metric_id": "personalized_alpha_ratio_accuracy", + "type": "numeric_check", + "target_key": "personalized_alpha_ratio", + "weight": 50, + "params": { + "gt_value": 26.436490698460958, + "tolerance": 1.321824534923048 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case35/case35_01.json b/foundational_analysis/cases/case35/case35_01.json new file mode 100644 index 0000000000000000000000000000000000000000..d2bd035565aacb0b55dddd7a4fc481b527728f67 --- /dev/null +++ b/foundational_analysis/cases/case35/case35_01.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 34.31, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 48.11, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 56.9, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_02.json b/foundational_analysis/cases/case35/case35_02.json new file mode 100644 index 0000000000000000000000000000000000000000..1df55cbdf1e35a9db1c3d6751949dd1a6b0b5032 --- /dev/null +++ b/foundational_analysis/cases/case35/case35_02.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 38.87, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 44.86, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 58.46, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_03.json b/foundational_analysis/cases/case35/case35_03.json new file mode 100644 index 0000000000000000000000000000000000000000..27c4628acca647028629a64ffa0341450089964d --- /dev/null +++ b/foundational_analysis/cases/case35/case35_03.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 39.97, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 44.7, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 57.55, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_04.json b/foundational_analysis/cases/case35/case35_04.json new file mode 100644 index 0000000000000000000000000000000000000000..a09ce8fdf5bf85ad899e2a1537f9d37058b2d873 --- /dev/null +++ b/foundational_analysis/cases/case35/case35_04.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 33.15, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 47.23, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 53.84, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_05.json b/foundational_analysis/cases/case35/case35_05.json new file mode 100644 index 0000000000000000000000000000000000000000..5c93653ded6b9dff7a6256ed97e75e79092dbff3 --- /dev/null +++ b/foundational_analysis/cases/case35/case35_05.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 36.89, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 45.83, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 59.99, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_06.json b/foundational_analysis/cases/case35/case35_06.json new file mode 100644 index 0000000000000000000000000000000000000000..643cb3f8cc082aa18c2089ea5fb7289b138a6f95 --- /dev/null +++ b/foundational_analysis/cases/case35/case35_06.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 33.88, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 46.77, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 52.63, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_07.json b/foundational_analysis/cases/case35/case35_07.json new file mode 100644 index 0000000000000000000000000000000000000000..2e6e9b22fa8d51020c2033ad519b071bc1282a06 --- /dev/null +++ b/foundational_analysis/cases/case35/case35_07.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 34.43, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 40.29, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 59.04, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_08.json b/foundational_analysis/cases/case35/case35_08.json new file mode 100644 index 0000000000000000000000000000000000000000..d56c40cf386cb7522c9a80491d58017079574332 --- /dev/null +++ b/foundational_analysis/cases/case35/case35_08.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 35.52, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 41.38, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 54.27, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_09.json b/foundational_analysis/cases/case35/case35_09.json new file mode 100644 index 0000000000000000000000000000000000000000..149bacfa47648a185f33fd4f942d51b9f3b3da88 --- /dev/null +++ b/foundational_analysis/cases/case35/case35_09.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 34.67, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 40.53, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 53.42, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_10.json b/foundational_analysis/cases/case35/case35_10.json new file mode 100644 index 0000000000000000000000000000000000000000..5eb4ac93f0bf89b65b25d87cb8ced386af4f6e99 --- /dev/null +++ b/foundational_analysis/cases/case35/case35_10.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 34.2, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 45.92, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 58.81, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_11.json b/foundational_analysis/cases/case35/case35_11.json new file mode 100644 index 0000000000000000000000000000000000000000..e41ecd667f06e24346a0f523a01c121c1975117a --- /dev/null +++ b/foundational_analysis/cases/case35/case35_11.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 36.96, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 42.17, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 51.76, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_12.json b/foundational_analysis/cases/case35/case35_12.json new file mode 100644 index 0000000000000000000000000000000000000000..9ef8b8a85218da95945119962609c7d27f26c26f --- /dev/null +++ b/foundational_analysis/cases/case35/case35_12.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 38.13, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 40.7, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 59.82, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_13.json b/foundational_analysis/cases/case35/case35_13.json new file mode 100644 index 0000000000000000000000000000000000000000..f9122d0eeb844c6e093b6b86589d0a6ca160a87a --- /dev/null +++ b/foundational_analysis/cases/case35/case35_13.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 34.04, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 40.85, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 58.23, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_14.json b/foundational_analysis/cases/case35/case35_14.json new file mode 100644 index 0000000000000000000000000000000000000000..58e3d2583a7fbb86864aa96399d83e1d6119f6ce --- /dev/null +++ b/foundational_analysis/cases/case35/case35_14.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 35.42, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 41.99, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 57.84, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_15.json b/foundational_analysis/cases/case35/case35_15.json new file mode 100644 index 0000000000000000000000000000000000000000..5f252e3f8c2bdc6956d161c04548b1185e2233ec --- /dev/null +++ b/foundational_analysis/cases/case35/case35_15.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 31.11, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 42.4, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 54.19, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_16.json b/foundational_analysis/cases/case35/case35_16.json new file mode 100644 index 0000000000000000000000000000000000000000..a0847131c2f57d6f22ebf44bf07e07d3a83c19cf --- /dev/null +++ b/foundational_analysis/cases/case35/case35_16.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 34.7, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 44.57, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 54.41, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_17.json b/foundational_analysis/cases/case35/case35_17.json new file mode 100644 index 0000000000000000000000000000000000000000..881f8d8943b1eedeff18c80da4e1ac3f84e34fd7 --- /dev/null +++ b/foundational_analysis/cases/case35/case35_17.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 33.91, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 46.87, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 50.43, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_18.json b/foundational_analysis/cases/case35/case35_18.json new file mode 100644 index 0000000000000000000000000000000000000000..95f619d97890dd16d82afb039ac57aee34b55e2c --- /dev/null +++ b/foundational_analysis/cases/case35/case35_18.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 33.87, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 43.14, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 51.14, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_19.json b/foundational_analysis/cases/case35/case35_19.json new file mode 100644 index 0000000000000000000000000000000000000000..5e8df8a381b9ddf642674763abcaf7e3d385652a --- /dev/null +++ b/foundational_analysis/cases/case35/case35_19.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 39.29, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 40.55, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 54.79, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_20.json b/foundational_analysis/cases/case35/case35_20.json new file mode 100644 index 0000000000000000000000000000000000000000..99833b6242d7cdd1eadea286cb9351a43b951c6e --- /dev/null +++ b/foundational_analysis/cases/case35/case35_20.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 32.89, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 49.56, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 54.21, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_21.json b/foundational_analysis/cases/case35/case35_21.json new file mode 100644 index 0000000000000000000000000000000000000000..a4cf97b91ca7e2588b51fd4162d7f05010689403 --- /dev/null +++ b/foundational_analysis/cases/case35/case35_21.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 35.92, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 46.83, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 52.96, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_22.json b/foundational_analysis/cases/case35/case35_22.json new file mode 100644 index 0000000000000000000000000000000000000000..d4bf3b3aab1a13c26bf8782556531d81ce73f872 --- /dev/null +++ b/foundational_analysis/cases/case35/case35_22.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 39.2, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 40.02, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 59.1, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_23.json b/foundational_analysis/cases/case35/case35_23.json new file mode 100644 index 0000000000000000000000000000000000000000..1c204852665ce7e01bf0ac1eb41afbd950d2d825 --- /dev/null +++ b/foundational_analysis/cases/case35/case35_23.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 35.04, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 42.85, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 54.23, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_24.json b/foundational_analysis/cases/case35/case35_24.json new file mode 100644 index 0000000000000000000000000000000000000000..72bf215f1bfa7a2ffec207f0b1585e8fed33f373 --- /dev/null +++ b/foundational_analysis/cases/case35/case35_24.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 35.1, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 48.28, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 53.67, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case35/case35_25.json b/foundational_analysis/cases/case35/case35_25.json new file mode 100644 index 0000000000000000000000000000000000000000..7bfe6942e6114894d6f9d804feefac3f604f418c --- /dev/null +++ b/foundational_analysis/cases/case35/case35_25.json @@ -0,0 +1,47 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please select standard EEG channels, apply common average reference, resample to 100Hz, apply a 0.5-40Hz FIR bandpass filter, and split the 30-60s segment into three 10-second windows. Compute Global Field Power (GFP) in each window and report the time point of the maximum GFP peak in each window." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the three GFP peak time points from the agent report:\n1) peak time in window 1\n2) peak time in window 2\n3) peak time in window 3\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"gfp_peak_time_window_1\", \"gfp_peak_time_window_2\", and \"gfp_peak_time_window_3\".\n5. Values must be float seconds or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"gfp_peak_time_window_1\": , \"gfp_peak_time_window_2\": , \"gfp_peak_time_window_3\": }", + "metrics": [ + { + "metric_id": "gfp_peak_time_window_1_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_1", + "weight": 34, + "params": { + "gt_value": 39.83, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_2_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_2", + "weight": 33, + "params": { + "gt_value": 46.62, + "tolerance": 0.1 + } + }, + { + "metric_id": "gfp_peak_time_window_3_accuracy", + "type": "numeric_check", + "target_key": "gfp_peak_time_window_3", + "weight": 33, + "params": { + "gt_value": 58.73, + "tolerance": 0.1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case36/case36_01.json b/foundational_analysis/cases/case36/case36_01.json new file mode 100644 index 0000000000000000000000000000000000000000..64fd79bf97674a05ab4311c73acf011b8ed9ee6d --- /dev/null +++ b/foundational_analysis/cases/case36/case36_01.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.10183837086706077, + "tolerance": 0.010183837086706077 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.11663708610077884, + "tolerance": 0.011663708610077886 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: right stronger. The true beta-band hemispheric pattern is: left stronger. The true cross-band hemispheric connectivity pattern is: not consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_02.json b/foundational_analysis/cases/case36/case36_02.json new file mode 100644 index 0000000000000000000000000000000000000000..c6d46692b2f9ae2698f25b2100678b917e9f38d9 --- /dev/null +++ b/foundational_analysis/cases/case36/case36_02.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.28287504654747525, + "tolerance": 0.028287504654747525 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.07796171232859583, + "tolerance": 0.0077961712328595834 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: right stronger. The true beta-band hemispheric pattern is: right stronger. The true cross-band hemispheric connectivity pattern is: consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_03.json b/foundational_analysis/cases/case36/case36_03.json new file mode 100644 index 0000000000000000000000000000000000000000..63d9a2b8dae9ad81576b0ea0e8fa2b0abe514d9d --- /dev/null +++ b/foundational_analysis/cases/case36/case36_03.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.12750544781538356, + "tolerance": 0.012750544781538357 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.0835590110723878, + "tolerance": 0.008355901107238781 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: right stronger. The true beta-band hemispheric pattern is: left stronger. The true cross-band hemispheric connectivity pattern is: not consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_04.json b/foundational_analysis/cases/case36/case36_04.json new file mode 100644 index 0000000000000000000000000000000000000000..36807325bf68c67b6f6913eaace8f4940c0cb9a8 --- /dev/null +++ b/foundational_analysis/cases/case36/case36_04.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.3561121593400625, + "tolerance": 0.035611215934006256 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.012752553420974477, + "tolerance": 0.0012752553420974479 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: right stronger. The true beta-band hemispheric pattern is: left stronger. The true cross-band hemispheric connectivity pattern is: not consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_05.json b/foundational_analysis/cases/case36/case36_05.json new file mode 100644 index 0000000000000000000000000000000000000000..8910655e838e5c21173861f87826e80944c2ddd6 --- /dev/null +++ b/foundational_analysis/cases/case36/case36_05.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.09959623877619658, + "tolerance": 0.0099596238776196583 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.02088723332267986, + "tolerance": 0.0020887233322679859 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: right stronger. The true beta-band hemispheric pattern is: left stronger. The true cross-band hemispheric connectivity pattern is: not consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_06.json b/foundational_analysis/cases/case36/case36_06.json new file mode 100644 index 0000000000000000000000000000000000000000..be56dad06ef4f18952f927930368d62d9d98521a --- /dev/null +++ b/foundational_analysis/cases/case36/case36_06.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.009856613109141155, + "tolerance": 0.0009856613109141155 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.02401876387634469, + "tolerance": 0.002401876387634469 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: left stronger. The true beta-band hemispheric pattern is: left stronger. The true cross-band hemispheric connectivity pattern is: consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_07.json b/foundational_analysis/cases/case36/case36_07.json new file mode 100644 index 0000000000000000000000000000000000000000..1e903e79f26015bdee5255534b552be5c0c76b1b --- /dev/null +++ b/foundational_analysis/cases/case36/case36_07.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.0647068520147116, + "tolerance": 0.00647068520147116 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.046912761488416055, + "tolerance": 0.004691276148841606 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: right stronger. The true beta-band hemispheric pattern is: left stronger. The true cross-band hemispheric connectivity pattern is: not consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_08.json b/foundational_analysis/cases/case36/case36_08.json new file mode 100644 index 0000000000000000000000000000000000000000..6ef356146e0fc6c65a1d88c9d019a4816a18012a --- /dev/null +++ b/foundational_analysis/cases/case36/case36_08.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.0011809606227857206, + "tolerance": 0.00011809606227857206 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.06397596852974191, + "tolerance": 0.0063975968529741915 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: left stronger. The true beta-band hemispheric pattern is: right stronger. The true cross-band hemispheric connectivity pattern is: not consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_09.json b/foundational_analysis/cases/case36/case36_09.json new file mode 100644 index 0000000000000000000000000000000000000000..e9ff1e91f420b7900bbda874252bc3de7f2e306a --- /dev/null +++ b/foundational_analysis/cases/case36/case36_09.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.04066328230934452, + "tolerance": 0.004066328230934452 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.034796697749316624, + "tolerance": 0.0034796697749316626 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: left stronger. The true beta-band hemispheric pattern is: right stronger. The true cross-band hemispheric connectivity pattern is: not consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_10.json b/foundational_analysis/cases/case36/case36_10.json new file mode 100644 index 0000000000000000000000000000000000000000..b79b6ca55e9a59c441b14ac9e6def01793048d42 --- /dev/null +++ b/foundational_analysis/cases/case36/case36_10.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.09007278773493277, + "tolerance": 0.009007278773493277 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.0036358729350781105, + "tolerance": 0.00036358729350781106 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: left stronger. The true beta-band hemispheric pattern is: right stronger. The true cross-band hemispheric connectivity pattern is: not consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_11.json b/foundational_analysis/cases/case36/case36_11.json new file mode 100644 index 0000000000000000000000000000000000000000..09a66747c8d203c6cdeb81e8e70b0f8c043f7ad7 --- /dev/null +++ b/foundational_analysis/cases/case36/case36_11.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.022098142473689714, + "tolerance": 0.0022098142473689713 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.040406653789793474, + "tolerance": 0.0040406653789793476 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: left stronger. The true beta-band hemispheric pattern is: left stronger. The true cross-band hemispheric connectivity pattern is: consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_12.json b/foundational_analysis/cases/case36/case36_12.json new file mode 100644 index 0000000000000000000000000000000000000000..af48eff72a4626ea0774adf86080bff4d3b51eba --- /dev/null +++ b/foundational_analysis/cases/case36/case36_12.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.16289535209135445, + "tolerance": 0.016289535209135446 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.005746606628711098, + "tolerance": 0.00057466066287110984 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: left stronger. The true beta-band hemispheric pattern is: right stronger. The true cross-band hemispheric connectivity pattern is: not consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_13.json b/foundational_analysis/cases/case36/case36_13.json new file mode 100644 index 0000000000000000000000000000000000000000..b80b629988c8d60e2760c9e954ca9d39a1ef85c8 --- /dev/null +++ b/foundational_analysis/cases/case36/case36_13.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.06392537357274522, + "tolerance": 0.0063925373572745227 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.024165981603348992, + "tolerance": 0.0024165981603348993 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: right stronger. The true beta-band hemispheric pattern is: right stronger. The true cross-band hemispheric connectivity pattern is: consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_14.json b/foundational_analysis/cases/case36/case36_14.json new file mode 100644 index 0000000000000000000000000000000000000000..815b423cccbfd62d97f6edd651216c1a54122294 --- /dev/null +++ b/foundational_analysis/cases/case36/case36_14.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.016689350354988183, + "tolerance": 0.0016689350354988184 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.01689309085754523, + "tolerance": 0.001689309085754523 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: left stronger. The true beta-band hemispheric pattern is: right stronger. The true cross-band hemispheric connectivity pattern is: not consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_15.json b/foundational_analysis/cases/case36/case36_15.json new file mode 100644 index 0000000000000000000000000000000000000000..b5a685b5837f73aef8f77a78e641d591b641a349 --- /dev/null +++ b/foundational_analysis/cases/case36/case36_15.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.10021942204774725, + "tolerance": 0.010021942204774726 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.041421996764194044, + "tolerance": 0.0041421996764194049 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: left stronger. The true beta-band hemispheric pattern is: left stronger. The true cross-band hemispheric connectivity pattern is: consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_16.json b/foundational_analysis/cases/case36/case36_16.json new file mode 100644 index 0000000000000000000000000000000000000000..bcbcf0ea88e47019dcd7e5ab6d5758d007a8ff1c --- /dev/null +++ b/foundational_analysis/cases/case36/case36_16.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.01536600411565281, + "tolerance": 0.0015366004115652809 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.013575376507554235, + "tolerance": 0.0013575376507554235 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: right stronger. The true beta-band hemispheric pattern is: right stronger. The true cross-band hemispheric connectivity pattern is: consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_17.json b/foundational_analysis/cases/case36/case36_17.json new file mode 100644 index 0000000000000000000000000000000000000000..5127f1baa1d84df35cde660f2ca1fde7828896b7 --- /dev/null +++ b/foundational_analysis/cases/case36/case36_17.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.1193149659109708, + "tolerance": 0.011931496591097081 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.02338608723753155, + "tolerance": 0.0023386087237531549 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: right stronger. The true beta-band hemispheric pattern is: left stronger. The true cross-band hemispheric connectivity pattern is: not consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_18.json b/foundational_analysis/cases/case36/case36_18.json new file mode 100644 index 0000000000000000000000000000000000000000..2d9c346a2cc67307e5d2edd52d2c7803829a3a10 --- /dev/null +++ b/foundational_analysis/cases/case36/case36_18.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.10295398424646471, + "tolerance": 0.010295398424646472 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.04024681947756084, + "tolerance": 0.0040246819477560841 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: left stronger. The true beta-band hemispheric pattern is: right stronger. The true cross-band hemispheric connectivity pattern is: not consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_19.json b/foundational_analysis/cases/case36/case36_19.json new file mode 100644 index 0000000000000000000000000000000000000000..829bf7671747a1810dfcf040c18775a1e8762588 --- /dev/null +++ b/foundational_analysis/cases/case36/case36_19.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.017871885964223788, + "tolerance": 0.0017871885964223788 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.03452771759220455, + "tolerance": 0.003452771759220455 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: right stronger. The true beta-band hemispheric pattern is: right stronger. The true cross-band hemispheric connectivity pattern is: consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_20.json b/foundational_analysis/cases/case36/case36_20.json new file mode 100644 index 0000000000000000000000000000000000000000..48fc9685344486209cf03a77accaec942afb89ca --- /dev/null +++ b/foundational_analysis/cases/case36/case36_20.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.16588329165386, + "tolerance": 0.016588329165386002 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.0008496848360674963, + "tolerance": 8.4968483606749634E-05 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: right stronger. The true beta-band hemispheric pattern is: left stronger. The true cross-band hemispheric connectivity pattern is: not consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_21.json b/foundational_analysis/cases/case36/case36_21.json new file mode 100644 index 0000000000000000000000000000000000000000..5f7ac7f6656ee3519ada28b8aa49ee03cfc821e4 --- /dev/null +++ b/foundational_analysis/cases/case36/case36_21.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.043898563887104114, + "tolerance": 0.0043898563887104112 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.02747075641605247, + "tolerance": 0.0027470756416052471 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: right stronger. The true beta-band hemispheric pattern is: left stronger. The true cross-band hemispheric connectivity pattern is: not consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_22.json b/foundational_analysis/cases/case36/case36_22.json new file mode 100644 index 0000000000000000000000000000000000000000..0430bc5f799807d84566fe0f33f569cb5a787ab6 --- /dev/null +++ b/foundational_analysis/cases/case36/case36_22.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.027913908378079832, + "tolerance": 0.0027913908378079835 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.030424366298207134, + "tolerance": 0.0030424366298207137 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: right stronger. The true beta-band hemispheric pattern is: left stronger. The true cross-band hemispheric connectivity pattern is: not consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_23.json b/foundational_analysis/cases/case36/case36_23.json new file mode 100644 index 0000000000000000000000000000000000000000..63f3496a4574361230126384019f37030d8bc304 --- /dev/null +++ b/foundational_analysis/cases/case36/case36_23.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.03579706207621729, + "tolerance": 0.003579706207621729 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.010030274365802463, + "tolerance": 0.0010030274365802464 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: left stronger. The true beta-band hemispheric pattern is: left stronger. The true cross-band hemispheric connectivity pattern is: consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_24.json b/foundational_analysis/cases/case36/case36_24.json new file mode 100644 index 0000000000000000000000000000000000000000..0605733f497598e6332ca4aeac896bd2a3d904b2 --- /dev/null +++ b/foundational_analysis/cases/case36/case36_24.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.058385915807636524, + "tolerance": 0.0058385915807636531 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": -0.01498032449253231, + "tolerance": 0.001498032449253231 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: right stronger. The true beta-band hemispheric pattern is: right stronger. The true cross-band hemispheric connectivity pattern is: consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case36/case36_25.json b/foundational_analysis/cases/case36/case36_25.json new file mode 100644 index 0000000000000000000000000000000000000000..543da8bbf23e8a5b6407fdb9da1f3833ef4e0b9f --- /dev/null +++ b/foundational_analysis/cases/case36/case36_25.json @@ -0,0 +1,46 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please use the first 60 seconds of the recording and select left-right homologous EEG channel pairs. In the alpha band and beta band separately, compute the left-hemisphere internal wPLI and right-hemisphere internal wPLI. Report the left-minus-right wPLI difference for alpha and beta, and explain whether the hemispheric connectivity pattern is consistent across the two bands." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two numeric values from the agent report:\n1) alpha-band left-minus-right wPLI difference\n2) beta-band left-minus-right wPLI difference\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. Keys must be EXACTLY \"alpha_wpli_diff\" and \"beta_wpli_diff\".\n5. Both values must be float or null.\n6. If a value is missing or cannot be found, use null for that key.\n\n### OUTPUT TEMPLATE\n{\"alpha_wpli_diff\": , \"beta_wpli_diff\": }", + "metrics": [ + { + "metric_id": "alpha_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "alpha_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.05787050834060395, + "tolerance": 0.0057870508340603956 + } + }, + { + "metric_id": "beta_wpli_diff_accuracy", + "type": "numeric_check", + "target_key": "beta_wpli_diff", + "weight": 30, + "params": { + "gt_value": 0.01896161712663305, + "tolerance": 0.0018961617126633053 + } + }, + { + "metric_id": "hemispheric_pattern_consistency", + "type": "semantic_consistency", + "target_key": "full_text", + "weight": 40, + "params": { + "judge_prompt": "You are judging semantic consistency for an EEG hemispheric-connectivity task. The true alpha-band hemispheric pattern is: left stronger. The true beta-band hemispheric pattern is: left stronger. The true cross-band hemispheric connectivity pattern is: consistent. Use these stated true conclusions as the ground truth; do not infer different conclusions. Mark the response as correct only if it correctly describes the stronger hemisphere in both bands and correctly states whether the two bands are consistent. Output only a valid JSON object with exactly one key named status and a boolean value True or False." + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_01.json b/foundational_analysis/cases/case37/case37_01.json new file mode 100644 index 0000000000000000000000000000000000000000..ddf811b1b41ccc7d761a681ce0986d48b556214d --- /dev/null +++ b/foundational_analysis/cases/case37/case37_01.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [ + "C3", + "O1", + "O2" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_02.json b/foundational_analysis/cases/case37/case37_02.json new file mode 100644 index 0000000000000000000000000000000000000000..b520d41af66d02a53a6e048d946b4306fadc7ebc --- /dev/null +++ b/foundational_analysis/cases/case37/case37_02.json @@ -0,0 +1,34 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [ + "C3", + "C4", + "F3", + "F4" + ], + "match_mode": "per_hit_fraction", + "top_k": 4, + "score_per_hit": 0.25 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_03.json b/foundational_analysis/cases/case37/case37_03.json new file mode 100644 index 0000000000000000000000000000000000000000..d72b1e766cff044ba8c6a95ad6e26f737cc7e0e6 --- /dev/null +++ b/foundational_analysis/cases/case37/case37_03.json @@ -0,0 +1,34 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [ + "C3", + "C4", + "O1", + "O2" + ], + "match_mode": "per_hit_fraction", + "top_k": 4, + "score_per_hit": 0.25 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_04.json b/foundational_analysis/cases/case37/case37_04.json new file mode 100644 index 0000000000000000000000000000000000000000..4a0302310918311093b0ef69247550746118c452 --- /dev/null +++ b/foundational_analysis/cases/case37/case37_04.json @@ -0,0 +1,34 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [ + "C3", + "C4", + "O1", + "O2" + ], + "match_mode": "per_hit_fraction", + "top_k": 4, + "score_per_hit": 0.25 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_05.json b/foundational_analysis/cases/case37/case37_05.json new file mode 100644 index 0000000000000000000000000000000000000000..fcc4f9a4d46f1b0a45e55341e159c483200769a0 --- /dev/null +++ b/foundational_analysis/cases/case37/case37_05.json @@ -0,0 +1,34 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [ + "C3", + "C4", + "F4", + "O2" + ], + "match_mode": "per_hit_fraction", + "top_k": 4, + "score_per_hit": 0.25 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_06.json b/foundational_analysis/cases/case37/case37_06.json new file mode 100644 index 0000000000000000000000000000000000000000..a9407b7da3c6231abd605357f67866ae3b83c755 --- /dev/null +++ b/foundational_analysis/cases/case37/case37_06.json @@ -0,0 +1,31 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [ + "P2" + ], + "match_mode": "per_hit_fraction", + "top_k": 1, + "score_per_hit": 1.0 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_07.json b/foundational_analysis/cases/case37/case37_07.json new file mode 100644 index 0000000000000000000000000000000000000000..c4371b33a5d68f7babce37a22dfac6220f3693c5 --- /dev/null +++ b/foundational_analysis/cases/case37/case37_07.json @@ -0,0 +1,28 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [], + "match_mode": "per_hit_fraction", + "top_k": 0 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_08.json b/foundational_analysis/cases/case37/case37_08.json new file mode 100644 index 0000000000000000000000000000000000000000..30bc67ce6990f9d84a6c311bfe67978e9be20101 --- /dev/null +++ b/foundational_analysis/cases/case37/case37_08.json @@ -0,0 +1,31 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [ + "PO3" + ], + "match_mode": "per_hit_fraction", + "top_k": 1, + "score_per_hit": 1.0 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_09.json b/foundational_analysis/cases/case37/case37_09.json new file mode 100644 index 0000000000000000000000000000000000000000..40bd93aed3c8edeeea1f111544e8df4f221a3912 --- /dev/null +++ b/foundational_analysis/cases/case37/case37_09.json @@ -0,0 +1,28 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [], + "match_mode": "per_hit_fraction", + "top_k": 0 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_10.json b/foundational_analysis/cases/case37/case37_10.json new file mode 100644 index 0000000000000000000000000000000000000000..f7acfa3a456c1e817840800d7bd90bd3692bab80 --- /dev/null +++ b/foundational_analysis/cases/case37/case37_10.json @@ -0,0 +1,28 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [], + "match_mode": "per_hit_fraction", + "top_k": 0 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_11.json b/foundational_analysis/cases/case37/case37_11.json new file mode 100644 index 0000000000000000000000000000000000000000..981680bb4eeda4d903dcf045d39461fc05502f2c --- /dev/null +++ b/foundational_analysis/cases/case37/case37_11.json @@ -0,0 +1,31 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [ + "CZ" + ], + "match_mode": "per_hit_fraction", + "top_k": 1, + "score_per_hit": 1.0 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_12.json b/foundational_analysis/cases/case37/case37_12.json new file mode 100644 index 0000000000000000000000000000000000000000..e116d63435c4852ba06c56fdb655a230615f5957 --- /dev/null +++ b/foundational_analysis/cases/case37/case37_12.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [ + "C2", + "CP1", + "CP2" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_13.json b/foundational_analysis/cases/case37/case37_13.json new file mode 100644 index 0000000000000000000000000000000000000000..691af68f17b2fde6399c6b2f4e7dfcb53cbd45ca --- /dev/null +++ b/foundational_analysis/cases/case37/case37_13.json @@ -0,0 +1,28 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [], + "match_mode": "per_hit_fraction", + "top_k": 0 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_14.json b/foundational_analysis/cases/case37/case37_14.json new file mode 100644 index 0000000000000000000000000000000000000000..905afab7df5b26eb9c67ab0445cf671111f65caa --- /dev/null +++ b/foundational_analysis/cases/case37/case37_14.json @@ -0,0 +1,28 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [], + "match_mode": "per_hit_fraction", + "top_k": 0 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_15.json b/foundational_analysis/cases/case37/case37_15.json new file mode 100644 index 0000000000000000000000000000000000000000..06189ffc04fdb48c3824252f247be3159c0fec7e --- /dev/null +++ b/foundational_analysis/cases/case37/case37_15.json @@ -0,0 +1,31 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [ + "PO3" + ], + "match_mode": "per_hit_fraction", + "top_k": 1, + "score_per_hit": 1.0 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_16.json b/foundational_analysis/cases/case37/case37_16.json new file mode 100644 index 0000000000000000000000000000000000000000..368496abf293eef85652ed025f29b50f5c92f0a2 --- /dev/null +++ b/foundational_analysis/cases/case37/case37_16.json @@ -0,0 +1,28 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [], + "match_mode": "per_hit_fraction", + "top_k": 0 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_17.json b/foundational_analysis/cases/case37/case37_17.json new file mode 100644 index 0000000000000000000000000000000000000000..45cd65e8f6c9e72035785803403394dde9b91f71 --- /dev/null +++ b/foundational_analysis/cases/case37/case37_17.json @@ -0,0 +1,33 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [ + "C4", + "CZ", + "F4" + ], + "match_mode": "per_hit_fraction", + "top_k": 3, + "score_per_hit": 0.3333333333333333 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_18.json b/foundational_analysis/cases/case37/case37_18.json new file mode 100644 index 0000000000000000000000000000000000000000..8e1b2fca9b4112d352c21d3bf81fff2fa2988081 --- /dev/null +++ b/foundational_analysis/cases/case37/case37_18.json @@ -0,0 +1,32 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [ + "F4", + "F8" + ], + "match_mode": "per_hit_fraction", + "top_k": 2, + "score_per_hit": 0.5 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_19.json b/foundational_analysis/cases/case37/case37_19.json new file mode 100644 index 0000000000000000000000000000000000000000..dee05c6d713b94d4b83334cab128e8afd0b8979b --- /dev/null +++ b/foundational_analysis/cases/case37/case37_19.json @@ -0,0 +1,31 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [ + "FP2" + ], + "match_mode": "per_hit_fraction", + "top_k": 1, + "score_per_hit": 1.0 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_20.json b/foundational_analysis/cases/case37/case37_20.json new file mode 100644 index 0000000000000000000000000000000000000000..42b7008b36acfa5de0d576944490eecbba91d5f8 --- /dev/null +++ b/foundational_analysis/cases/case37/case37_20.json @@ -0,0 +1,28 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [], + "match_mode": "per_hit_fraction", + "top_k": 0 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_21.json b/foundational_analysis/cases/case37/case37_21.json new file mode 100644 index 0000000000000000000000000000000000000000..ade8ad7297f06b7a5e9f2abbf8b453ea8c125353 --- /dev/null +++ b/foundational_analysis/cases/case37/case37_21.json @@ -0,0 +1,32 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [ + "C4", + "CZ" + ], + "match_mode": "per_hit_fraction", + "top_k": 2, + "score_per_hit": 0.5 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_22.json b/foundational_analysis/cases/case37/case37_22.json new file mode 100644 index 0000000000000000000000000000000000000000..eb5e6a8537841ab7c5a36e2bd4f058394e6206bc --- /dev/null +++ b/foundational_analysis/cases/case37/case37_22.json @@ -0,0 +1,28 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [], + "match_mode": "per_hit_fraction", + "top_k": 0 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_23.json b/foundational_analysis/cases/case37/case37_23.json new file mode 100644 index 0000000000000000000000000000000000000000..cb856c1217c126244465cd132e69e064275725e1 --- /dev/null +++ b/foundational_analysis/cases/case37/case37_23.json @@ -0,0 +1,32 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [ + "CZ", + "F8" + ], + "match_mode": "per_hit_fraction", + "top_k": 2, + "score_per_hit": 0.5 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_24.json b/foundational_analysis/cases/case37/case37_24.json new file mode 100644 index 0000000000000000000000000000000000000000..a367c843fc89f8af5d710378277cae4f8b4af2d8 --- /dev/null +++ b/foundational_analysis/cases/case37/case37_24.json @@ -0,0 +1,28 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [], + "match_mode": "per_hit_fraction", + "top_k": 0 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case37/case37_25.json b/foundational_analysis/cases/case37/case37_25.json new file mode 100644 index 0000000000000000000000000000000000000000..27749cac210e953cfeaeb6f9c793b01109cca2f6 --- /dev/null +++ b/foundational_analysis/cases/case37/case37_25.json @@ -0,0 +1,31 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 1.5, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}", + "metrics": [ + { + "metric_id": "multi_band_top_channel_set", + "type": "set_match_check", + "target_key": "multi_band_top_channels", + "weight": 100, + "params": { + "gt_value": [ + "FZ" + ], + "match_mode": "per_hit_fraction", + "top_k": 1, + "score_per_hit": 1.0 + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case38/case38_01.json b/foundational_analysis/cases/case38/case38_01.json new file mode 100644 index 0000000000000000000000000000000000000000..18066410ba3f396d892028d7ab000bbdc416b9b6 --- /dev/null +++ b/foundational_analysis/cases/case38/case38_01.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "F4", + "C4", + "F3", + "C3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_02.json b/foundational_analysis/cases/case38/case38_02.json new file mode 100644 index 0000000000000000000000000000000000000000..49544d972112c79609616c2c19c0010d0f3d4b92 --- /dev/null +++ b/foundational_analysis/cases/case38/case38_02.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "C4", + "C3", + "F4", + "F3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_03.json b/foundational_analysis/cases/case38/case38_03.json new file mode 100644 index 0000000000000000000000000000000000000000..e08a67182b5b99e618dcdb04b2c85878a3503f5a --- /dev/null +++ b/foundational_analysis/cases/case38/case38_03.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "F4", + "F3", + "C3", + "C4" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_04.json b/foundational_analysis/cases/case38/case38_04.json new file mode 100644 index 0000000000000000000000000000000000000000..e10520596918007a25446e7a31659e2982fb0e99 --- /dev/null +++ b/foundational_analysis/cases/case38/case38_04.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "C4", + "F4", + "F3", + "C3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_05.json b/foundational_analysis/cases/case38/case38_05.json new file mode 100644 index 0000000000000000000000000000000000000000..9fe5c438c93a8c9b6cb6fb957b386e84347d6623 --- /dev/null +++ b/foundational_analysis/cases/case38/case38_05.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "C4", + "F4", + "C3", + "F3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_06.json b/foundational_analysis/cases/case38/case38_06.json new file mode 100644 index 0000000000000000000000000000000000000000..b75e0354e72d40bc7338bd143ecb1113e3c59422 --- /dev/null +++ b/foundational_analysis/cases/case38/case38_06.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "AF7", + "AF8", + "F5", + "F7" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_07.json b/foundational_analysis/cases/case38/case38_07.json new file mode 100644 index 0000000000000000000000000000000000000000..236914ae7f2f0f97a6ec9a7152385e32986621d9 --- /dev/null +++ b/foundational_analysis/cases/case38/case38_07.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "AF4", + "F3", + "AF3", + "FZ" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_08.json b/foundational_analysis/cases/case38/case38_08.json new file mode 100644 index 0000000000000000000000000000000000000000..6c06fce0f94be8512e8485114661d6245c8153ac --- /dev/null +++ b/foundational_analysis/cases/case38/case38_08.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "AF3", + "FP2", + "FP1", + "AF4" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_09.json b/foundational_analysis/cases/case38/case38_09.json new file mode 100644 index 0000000000000000000000000000000000000000..9e349ffe36328baee8f4a8f8ab0b725cccf7c50e --- /dev/null +++ b/foundational_analysis/cases/case38/case38_09.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "FP1", + "AF4", + "F5", + "F3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_10.json b/foundational_analysis/cases/case38/case38_10.json new file mode 100644 index 0000000000000000000000000000000000000000..36944e54736f0d37ca162a8e04e018a178324ef8 --- /dev/null +++ b/foundational_analysis/cases/case38/case38_10.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "AF4", + "F5", + "AF3", + "FZ" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_11.json b/foundational_analysis/cases/case38/case38_11.json new file mode 100644 index 0000000000000000000000000000000000000000..66f4b235b12bc5b6aa4d087e3ff23a86b6c25814 --- /dev/null +++ b/foundational_analysis/cases/case38/case38_11.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "AF4", + "FPZ", + "FP2", + "AF3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_12.json b/foundational_analysis/cases/case38/case38_12.json new file mode 100644 index 0000000000000000000000000000000000000000..231f578803d2bb6cedcc75bfc3704b61a291d2cf --- /dev/null +++ b/foundational_analysis/cases/case38/case38_12.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "AF3", + "FPZ", + "FP1", + "AF4" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_13.json b/foundational_analysis/cases/case38/case38_13.json new file mode 100644 index 0000000000000000000000000000000000000000..efaae117a668ae076a3ebb7b982df0f53ecf3946 --- /dev/null +++ b/foundational_analysis/cases/case38/case38_13.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "AF3", + "AF4", + "FPZ", + "F1" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_14.json b/foundational_analysis/cases/case38/case38_14.json new file mode 100644 index 0000000000000000000000000000000000000000..29cd7eb8cadffaa6b0a4ae3fafe1148577f642c9 --- /dev/null +++ b/foundational_analysis/cases/case38/case38_14.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "AF4", + "AF3", + "FZ", + "F2" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_15.json b/foundational_analysis/cases/case38/case38_15.json new file mode 100644 index 0000000000000000000000000000000000000000..7163aff285842ae40a96f68ed42a6b2e7521443f --- /dev/null +++ b/foundational_analysis/cases/case38/case38_15.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "FPZ", + "AF4", + "AF3", + "FP2" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_16.json b/foundational_analysis/cases/case38/case38_16.json new file mode 100644 index 0000000000000000000000000000000000000000..aa1ec6a8dd5478009804f42989e918ebfa0d4f26 --- /dev/null +++ b/foundational_analysis/cases/case38/case38_16.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "CZ", + "PZ", + "P3", + "F3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_17.json b/foundational_analysis/cases/case38/case38_17.json new file mode 100644 index 0000000000000000000000000000000000000000..59808cf03d3004bf50cf4f90ec88c31a75ac7b98 --- /dev/null +++ b/foundational_analysis/cases/case38/case38_17.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "F4", + "F3", + "FZ", + "FP1" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_18.json b/foundational_analysis/cases/case38/case38_18.json new file mode 100644 index 0000000000000000000000000000000000000000..dbabec3e5479aee606152411ced222b145f0f5ed --- /dev/null +++ b/foundational_analysis/cases/case38/case38_18.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "CZ", + "FZ", + "PZ", + "F4" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_19.json b/foundational_analysis/cases/case38/case38_19.json new file mode 100644 index 0000000000000000000000000000000000000000..92a3dfd611a3b9fb519b7c543f47232184a7634b --- /dev/null +++ b/foundational_analysis/cases/case38/case38_19.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "O1", + "O2", + "F3", + "FZ" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_20.json b/foundational_analysis/cases/case38/case38_20.json new file mode 100644 index 0000000000000000000000000000000000000000..90103e09a0b5b6a7c4a9b5a487cf79fa99277ef3 --- /dev/null +++ b/foundational_analysis/cases/case38/case38_20.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "F3", + "F4", + "FP2", + "FP1" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_21.json b/foundational_analysis/cases/case38/case38_21.json new file mode 100644 index 0000000000000000000000000000000000000000..7ad14be9b957be1f1361dcb40aa090a61fb344bf --- /dev/null +++ b/foundational_analysis/cases/case38/case38_21.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "FZ", + "F4", + "FP1", + "FP2" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_22.json b/foundational_analysis/cases/case38/case38_22.json new file mode 100644 index 0000000000000000000000000000000000000000..0a3b589593ec5d18e427fef7b5ac3b9fd1abdc70 --- /dev/null +++ b/foundational_analysis/cases/case38/case38_22.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "CZ", + "C3", + "C4", + "F3" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_23.json b/foundational_analysis/cases/case38/case38_23.json new file mode 100644 index 0000000000000000000000000000000000000000..a065ff76294631ec6cca50a01f1bd8bf2990ca83 --- /dev/null +++ b/foundational_analysis/cases/case38/case38_23.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "FP2", + "C3", + "CZ", + "C4" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_24.json b/foundational_analysis/cases/case38/case38_24.json new file mode 100644 index 0000000000000000000000000000000000000000..6b942ab331b48d9f4926de7ead5f618761dedce2 --- /dev/null +++ b/foundational_analysis/cases/case38/case38_24.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "C4", + "C3", + "F3", + "FZ" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case38/case38_25.json b/foundational_analysis/cases/case38/case38_25.json new file mode 100644 index 0000000000000000000000000000000000000000..61f2b310fbe55313f6a995335b0e55a3413fba48 --- /dev/null +++ b/foundational_analysis/cases/case38/case38_25.json @@ -0,0 +1,41 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please select standard EEG channels. For BCIC2020-3 and SEED-V recordings, restrict the analysis to frontal-region EEG channels; for other datasets, use all standard EEG channels. Apply an 8-13Hz zero-phase FIR bandpass filter to the selected continuous signals, then use the first 60 seconds of the filtered data. Extract instantaneous phase with the Hilbert transform and compute PLV for every unique channel pair. Define each channel's degree as the sum of its PLV connection strengths, and output the top-4 channel names ranked by degree from highest to lowest." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the ranked top-4 EEG channel names by PLV degree.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"plv_degree_top4_channels\".\n5. Extract the final ranked list, not a channel inventory or an intermediate PLV table. The value must be an array of exactly 4 strings.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each output label must contain only uppercase A-Z letters and digits. Examples: F4-A1 -> F4; EEG Fp1-LE -> FP1; EEG T4 -> T8; Cz -> CZ.\n10. Keep the four channels in descending PLV degree order. Do not sort the array alphabetically and do not substitute different channels.\n11. If the final list does not contain exactly 4 identifiable channel labels, return {\"plv_degree_top4_channels\": null}; do not guess.\n\n### OUTPUT TEMPLATE\n{\"plv_degree_top4_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\"]}", + "metrics": [ + { + "metric_id": "plv_degree_top4_sequence", + "type": "sequence_match_check", + "target_key": "plv_degree_top4_channels", + "weight": 100, + "params": { + "gt_value": [ + "CZ", + "C4", + "C3", + "T8" + ], + "match_mode": "weighted_partial_order", + "top_k": 4, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4 + ], + "min_overlap": 2, + "allow_order_slip": 1 + } + } + ] + } +} diff --git a/foundational_analysis/cases/case39/case39_01.json b/foundational_analysis/cases/case39/case39_01.json new file mode 100644 index 0000000000000000000000000000000000000000..c80e776341519d141effaf7689b75171e93c3d5d --- /dev/null +++ b/foundational_analysis/cases/case39/case39_01.json @@ -0,0 +1,56 @@ +{ + "meta_info": { + "case_id": "ISRUC_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_01.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[F3-A2, F4-A1]; central=[C3-A2, C4-A1]; parietal=[]; occipital=[O1-A2, O2-A1]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "C3", + "F4", + "O2", + "C4", + "O1" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "central", + "occipital" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_02.json b/foundational_analysis/cases/case39/case39_02.json new file mode 100644 index 0000000000000000000000000000000000000000..e90c2404f5f6741b2c9fed72a5381424551a5c0f --- /dev/null +++ b/foundational_analysis/cases/case39/case39_02.json @@ -0,0 +1,56 @@ +{ + "meta_info": { + "case_id": "ISRUC_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_02.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[F3-A2, F4-A1]; central=[C3-A2, C4-A1]; parietal=[]; occipital=[O1-A2, O2-A1]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "F3", + "F4", + "C3", + "C4", + "O1" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "frontal", + "central" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_03.json b/foundational_analysis/cases/case39/case39_03.json new file mode 100644 index 0000000000000000000000000000000000000000..a1bcfdd5e8758ffd37c6c8a93cb5df9efb892c7e --- /dev/null +++ b/foundational_analysis/cases/case39/case39_03.json @@ -0,0 +1,56 @@ +{ + "meta_info": { + "case_id": "ISRUC_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_03.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[F3-A2, F4-A1]; central=[C3-A2, C4-A1]; parietal=[]; occipital=[O1-A2, O2-A1]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "F4", + "F3", + "O2", + "O1", + "C4" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "frontal", + "occipital" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_04.json b/foundational_analysis/cases/case39/case39_04.json new file mode 100644 index 0000000000000000000000000000000000000000..8c4f5fec3cf26bbfbe3a54fdef0ded45f5d72591 --- /dev/null +++ b/foundational_analysis/cases/case39/case39_04.json @@ -0,0 +1,56 @@ +{ + "meta_info": { + "case_id": "ISRUC_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_04.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[F3-A2, F4-A1]; central=[C3-A2, C4-A1]; parietal=[]; occipital=[O1-A2, O2-A1]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "F3", + "F4", + "C3", + "C4", + "O1" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "frontal", + "central" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_05.json b/foundational_analysis/cases/case39/case39_05.json new file mode 100644 index 0000000000000000000000000000000000000000..66757ce9fd80bfe5d785af3cb3e71ee7b6325437 --- /dev/null +++ b/foundational_analysis/cases/case39/case39_05.json @@ -0,0 +1,56 @@ +{ + "meta_info": { + "case_id": "ISRUC_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "ISRUC" + }, + "agent_input": { + "data_path": "data/core/ISRUC_05.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[F3-A2, F4-A1]; central=[C3-A2, C4-A1]; parietal=[]; occipital=[O1-A2, O2-A1]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "F4", + "F3", + "C3", + "C4", + "O1" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "frontal", + "central" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_06.json b/foundational_analysis/cases/case39/case39_06.json new file mode 100644 index 0000000000000000000000000000000000000000..c86893d87bd3fb88bff930348ccdfdec3bb4e0d6 --- /dev/null +++ b/foundational_analysis/cases/case39/case39_06.json @@ -0,0 +1,55 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_01.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[Fp1, Fp2, AF7, AF3, AF4, AF8, F7, F5, F3, F1, Fz, F2, F4, F6, F8]; central=[FC5, FC1, FC2, FC6, T7, C3, Cz, C4, T8, FT9, FT7, FC3, FC4, FT8, FT10, C5, C1, C2, C6, CP3, CPz, CP4]; parietal=[CP5, CP1, CP2, CP6, TP9, TP10, TP7, TP8, P7, P3, Pz, P4, P8, P5, P1, P2, P6]; occipital=[PO9, O1, Oz, O2, PO10, PO7, PO3, POz, PO4, PO8]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "AF7", + "AF8", + "FP2", + "PO9", + "F7" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "frontal" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_07.json b/foundational_analysis/cases/case39/case39_07.json new file mode 100644 index 0000000000000000000000000000000000000000..a1774e83761b4b89e01d9b8aea358360a2dccebf --- /dev/null +++ b/foundational_analysis/cases/case39/case39_07.json @@ -0,0 +1,55 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_02.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[Fp1, Fp2, AF7, AF3, AF4, AF8, F7, F5, F3, F1, Fz, F2, F4, F6, F8]; central=[FC5, FC1, FC2, FC6, T7, C3, Cz, C4, T8, FT9, FT7, FC3, FC4, FT8, FT10, C5, C1, C2, C6, CP3, CPz, CP4]; parietal=[CP5, CP1, CP2, CP6, TP9, TP10, TP7, TP8, P7, P3, Pz, P4, P8, P5, P1, P2, P6]; occipital=[PO9, O1, Oz, O2, PO10, PO7, PO3, POz, PO4, PO8]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "FP2", + "FP1", + "F6", + "AF7", + "F8" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "frontal" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_08.json b/foundational_analysis/cases/case39/case39_08.json new file mode 100644 index 0000000000000000000000000000000000000000..251285eb951196e44387572d6ad627094c02a73b --- /dev/null +++ b/foundational_analysis/cases/case39/case39_08.json @@ -0,0 +1,55 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_03.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[Fp1, Fp2, AF7, AF3, AF4, AF8, F7, F5, F3, F1, Fz, F2, F4, F6, F8]; central=[FC5, FC1, FC2, FC6, T7, C3, Cz, C4, T8, FT9, FT7, FC3, FC4, FT8, FT10, C5, C1, C2, C6, CP3, CPz, CP4]; parietal=[CP5, CP1, CP2, CP6, TP9, TP10, TP7, TP8, P7, P3, Pz, P4, P8, P5, P1, P2, P6]; occipital=[PO9, O1, Oz, O2, PO10, PO7, PO3, POz, PO4, PO8]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "AF8", + "FP2", + "AF7", + "FP1", + "F7" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "frontal" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_09.json b/foundational_analysis/cases/case39/case39_09.json new file mode 100644 index 0000000000000000000000000000000000000000..f092d417152a7f47331f0c714c04bdcf1c8847a5 --- /dev/null +++ b/foundational_analysis/cases/case39/case39_09.json @@ -0,0 +1,55 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_04.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[Fp1, Fp2, AF7, AF3, AF4, AF8, F7, F5, F3, F1, Fz, F2, F4, F6, F8]; central=[FC5, FC1, FC2, FC6, T7, C3, Cz, C4, T8, FT9, FT7, FC3, FC4, FT8, FT10, C5, C1, C2, C6, CP3, CPz, CP4]; parietal=[CP5, CP1, CP2, CP6, TP9, TP10, TP7, TP8, P7, P3, Pz, P4, P8, P5, P1, P2, P6]; occipital=[PO9, O1, Oz, O2, PO10, PO7, PO3, POz, PO4, PO8]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "FT8", + "F7", + "T7", + "FT7", + "F8" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "central" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_10.json b/foundational_analysis/cases/case39/case39_10.json new file mode 100644 index 0000000000000000000000000000000000000000..941837b652dd2f973acd557bf6e365e1ae08ee03 --- /dev/null +++ b/foundational_analysis/cases/case39/case39_10.json @@ -0,0 +1,55 @@ +{ + "meta_info": { + "case_id": "BCIC2020-3_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "BCIC2020-3" + }, + "agent_input": { + "data_path": "data/core/BCIC2020-3_05.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[Fp1, Fp2, AF7, AF3, AF4, AF8, F7, F5, F3, F1, Fz, F2, F4, F6, F8]; central=[FC5, FC1, FC2, FC6, T7, C3, Cz, C4, T8, FT9, FT7, FC3, FC4, FT8, FT10, C5, C1, C2, C6, CP3, CPz, CP4]; parietal=[CP5, CP1, CP2, CP6, TP9, TP10, TP7, TP8, P7, P3, Pz, P4, P8, P5, P1, P2, P6]; occipital=[PO9, O1, Oz, O2, PO10, PO7, PO3, POz, PO4, PO8]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "AF7", + "TP7", + "F7", + "FP1", + "TP9" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "frontal" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_11.json b/foundational_analysis/cases/case39/case39_11.json new file mode 100644 index 0000000000000000000000000000000000000000..383eac45f8013f6d48cf7d2cae32e3a13efa2df4 --- /dev/null +++ b/foundational_analysis/cases/case39/case39_11.json @@ -0,0 +1,55 @@ +{ + "meta_info": { + "case_id": "SEED-V-01.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-01.cnt", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[FP1, FPZ, FP2, AF3, AF4, F7, F5, F3, F1, FZ, F2, F4, F6, F8]; central=[FT7, FC5, FC3, FC1, FCZ, FC2, FC4, FC6, FT8, T7, C5, C3, C1, CZ, C2, C4, C6, T8, CP3, CPZ, CP4]; parietal=[TP7, CP5, CP3, CP1, CPZ, CP2, CP4, CP6, TP8, P7, P5, P3, P1, PZ, P2, P4, P6, P8]; occipital=[PO7, PO5, PO3, POZ, PO4, PO6, PO8, O1, OZ, O2, CB1, CB2]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "FP1", + "FPZ", + "FP2", + "AF3", + "T7" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "frontal" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_12.json b/foundational_analysis/cases/case39/case39_12.json new file mode 100644 index 0000000000000000000000000000000000000000..d75465cbbdf65d73d9ea3f40eefe5e6b641fd4a6 --- /dev/null +++ b/foundational_analysis/cases/case39/case39_12.json @@ -0,0 +1,55 @@ +{ + "meta_info": { + "case_id": "SEED-V-02.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-02.cnt", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[FP1, FPZ, FP2, AF3, AF4, F7, F5, F3, F1, FZ, F2, F4, F6, F8]; central=[FT7, FC5, FC3, FC1, FCZ, FC2, FC4, FC6, FT8, T7, C5, C3, C1, CZ, C2, C4, C6, T8, CP3, CPZ, CP4]; parietal=[TP7, CP5, CP3, CP1, CPZ, CP2, CP4, CP6, TP8, P7, P5, P3, P1, PZ, P2, P4, P6, P8]; occipital=[PO7, PO5, PO3, POZ, PO4, PO6, PO8, O1, OZ, O2, CB1, CB2]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "FP1", + "FPZ", + "FP2", + "CB1", + "AF4" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "frontal" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_13.json b/foundational_analysis/cases/case39/case39_13.json new file mode 100644 index 0000000000000000000000000000000000000000..1f943b3ccd739df0957e9b65cfe8e3a12c496a42 --- /dev/null +++ b/foundational_analysis/cases/case39/case39_13.json @@ -0,0 +1,56 @@ +{ + "meta_info": { + "case_id": "SEED-V-03.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-03.cnt", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[FP1, FPZ, FP2, AF3, AF4, F7, F5, F3, F1, FZ, F2, F4, F6, F8]; central=[FT7, FC5, FC3, FC1, FCZ, FC2, FC4, FC6, FT8, T7, C5, C3, C1, CZ, C2, C4, C6, T8, CP3, CPZ, CP4]; parietal=[TP7, CP5, CP3, CP1, CPZ, CP2, CP4, CP6, TP8, P7, P5, P3, P1, PZ, P2, P4, P6, P8]; occipital=[PO7, PO5, PO3, POZ, PO4, PO6, PO8, O1, OZ, O2, CB1, CB2]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "F2", + "F4", + "C2", + "CB1", + "CB2" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "frontal", + "occipital" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_14.json b/foundational_analysis/cases/case39/case39_14.json new file mode 100644 index 0000000000000000000000000000000000000000..cbec95ce368fe3a3ab5c4523e77fd50209f8e83e --- /dev/null +++ b/foundational_analysis/cases/case39/case39_14.json @@ -0,0 +1,55 @@ +{ + "meta_info": { + "case_id": "SEED-V-04.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-04.cnt", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[FP1, FPZ, FP2, AF3, AF4, F7, F5, F3, F1, FZ, F2, F4, F6, F8]; central=[FT7, FC5, FC3, FC1, FCZ, FC2, FC4, FC6, FT8, T7, C5, C3, C1, CZ, C2, C4, C6, T8, CP3, CPZ, CP4]; parietal=[TP7, CP5, CP3, CP1, CPZ, CP2, CP4, CP6, TP8, P7, P5, P3, P1, PZ, P2, P4, P6, P8]; occipital=[PO7, PO5, PO3, POZ, PO4, PO6, PO8, O1, OZ, O2, CB1, CB2]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "T7", + "PO7", + "P7", + "CB1", + "O1" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "occipital" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_15.json b/foundational_analysis/cases/case39/case39_15.json new file mode 100644 index 0000000000000000000000000000000000000000..0551ed3e5d0ea0cc729c9f98494135b801c8aaaf --- /dev/null +++ b/foundational_analysis/cases/case39/case39_15.json @@ -0,0 +1,55 @@ +{ + "meta_info": { + "case_id": "SEED-V-05.cnt", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "SEED-V" + }, + "agent_input": { + "data_path": "data/core/SEED-V-05.cnt", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[FP1, FPZ, FP2, AF3, AF4, F7, F5, F3, F1, FZ, F2, F4, F6, F8]; central=[FT7, FC5, FC3, FC1, FCZ, FC2, FC4, FC6, FT8, T7, C5, C3, C1, CZ, C2, C4, C6, T8, CP3, CPZ, CP4]; parietal=[TP7, CP5, CP3, CP1, CPZ, CP2, CP4, CP6, TP8, P7, P5, P3, P1, PZ, P2, P4, P6, P8]; occipital=[PO7, PO5, PO3, POZ, PO4, PO6, PO8, O1, OZ, O2, CB1, CB2]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "PO5", + "PO3", + "O1", + "CB1", + "PO7" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "occipital" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_16.json b/foundational_analysis/cases/case39/case39_16.json new file mode 100644 index 0000000000000000000000000000000000000000..054b895dfb6e2f099a12fd3a4eac57c850eaa9ae --- /dev/null +++ b/foundational_analysis/cases/case39/case39_16.json @@ -0,0 +1,55 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_01.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[Fp1-LE, F3-LE, F7-LE, Fz-LE, Fp2-LE, F4-LE, F8-LE]; central=[C3-LE, C4-LE, Cz-LE]; parietal=[P3-LE, P4-LE, Pz-LE, T5-LE, T6-LE]; occipital=[O1-LE, O2-LE]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "P3", + "PZ", + "P4", + "O2", + "C3" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "parietal" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_17.json b/foundational_analysis/cases/case39/case39_17.json new file mode 100644 index 0000000000000000000000000000000000000000..1f001c4c43c70015e24beb6eca039c74bbd4bf5e --- /dev/null +++ b/foundational_analysis/cases/case39/case39_17.json @@ -0,0 +1,55 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_02.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[Fp1-LE, F3-LE, F7-LE, Fz-LE, Fp2-LE, F4-LE, F8-LE]; central=[C3-LE, C4-LE, Cz-LE]; parietal=[P3-LE, P4-LE, Pz-LE, T5-LE, T6-LE]; occipital=[O1-LE, O2-LE]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "O2", + "O1", + "P4", + "PZ", + "P3" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "parietal" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_18.json b/foundational_analysis/cases/case39/case39_18.json new file mode 100644 index 0000000000000000000000000000000000000000..bb335487039ddcf84b1e3307d741852b8ec356b8 --- /dev/null +++ b/foundational_analysis/cases/case39/case39_18.json @@ -0,0 +1,56 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_03.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[Fp1-LE, F3-LE, F7-LE, Fz-LE, Fp2-LE, F4-LE, F8-LE]; central=[C3-LE, C4-LE, Cz-LE]; parietal=[P3-LE, P4-LE, Pz-LE, T5-LE, T6-LE]; occipital=[O1-LE, O2-LE]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "O2", + "O1", + "PZ", + "CZ", + "P8" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "parietal", + "occipital" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_19.json b/foundational_analysis/cases/case39/case39_19.json new file mode 100644 index 0000000000000000000000000000000000000000..c8e480751c4f120565c4e0d128a05f412530795a --- /dev/null +++ b/foundational_analysis/cases/case39/case39_19.json @@ -0,0 +1,55 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_04.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[Fp1-LE, F3-LE, F7-LE, Fz-LE, Fp2-LE, F4-LE, F8-LE]; central=[C3-LE, C4-LE, Cz-LE]; parietal=[P3-LE, P4-LE, Pz-LE, T5-LE, T6-LE]; occipital=[O1-LE, O2-LE]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "O2", + "O1", + "P7", + "P3", + "PZ" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "parietal" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_20.json b/foundational_analysis/cases/case39/case39_20.json new file mode 100644 index 0000000000000000000000000000000000000000..b3ae9396b7b4b403a64a12d1e163dc9c22dfcd45 --- /dev/null +++ b/foundational_analysis/cases/case39/case39_20.json @@ -0,0 +1,55 @@ +{ + "meta_info": { + "case_id": "Mumtaz2016_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "Mumtaz2016" + }, + "agent_input": { + "data_path": "data/core/Mumtaz2016_05.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[Fp1-LE, F3-LE, F7-LE, Fz-LE, Fp2-LE, F4-LE, F8-LE]; central=[C3-LE, C4-LE, Cz-LE]; parietal=[P3-LE, P4-LE, Pz-LE, T5-LE, T6-LE]; occipital=[O1-LE, O2-LE]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "P8", + "O1", + "O2", + "P4", + "P3" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "parietal" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_21.json b/foundational_analysis/cases/case39/case39_21.json new file mode 100644 index 0000000000000000000000000000000000000000..a36d47b3e4648e0bfc8521c73c82f072777baacd --- /dev/null +++ b/foundational_analysis/cases/case39/case39_21.json @@ -0,0 +1,55 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_01.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_01.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[Fp1, Fp2, F3, F4, F7, F8, Fz]; central=[C3, C4, Cz, T3, T4]; parietal=[P3, P4, Pz, T5, T6]; occipital=[O1, O2]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "O2", + "P4", + "P8", + "O1", + "PZ" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "parietal" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_22.json b/foundational_analysis/cases/case39/case39_22.json new file mode 100644 index 0000000000000000000000000000000000000000..2a155f096695b1ba7534a49f5eb7298879d0be6e --- /dev/null +++ b/foundational_analysis/cases/case39/case39_22.json @@ -0,0 +1,55 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_02.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_02.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[Fp1, Fp2, F3, F4, F7, F8, Fz]; central=[C3, C4, Cz, T3, T4]; parietal=[P3, P4, Pz, T5, T6]; occipital=[O1, O2]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "O2", + "O1", + "P4", + "PZ", + "P3" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "parietal" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_23.json b/foundational_analysis/cases/case39/case39_23.json new file mode 100644 index 0000000000000000000000000000000000000000..c99766943df0bcd95f1f9671993d86cd5390a8fc --- /dev/null +++ b/foundational_analysis/cases/case39/case39_23.json @@ -0,0 +1,56 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_03.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_03.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[Fp1, Fp2, F3, F4, F7, F8, Fz]; central=[C3, C4, Cz, T3, T4]; parietal=[P3, P4, Pz, T5, T6]; occipital=[O1, O2]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "O2", + "O1", + "FP2", + "P4", + "PZ" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "parietal", + "occipital" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_24.json b/foundational_analysis/cases/case39/case39_24.json new file mode 100644 index 0000000000000000000000000000000000000000..f86da5adb7737eaa146ff9e5b854fa691dcff44d --- /dev/null +++ b/foundational_analysis/cases/case39/case39_24.json @@ -0,0 +1,55 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_04.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_04.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[Fp1, Fp2, F3, F4, F7, F8, Fz]; central=[C3, C4, Cz, T3, T4]; parietal=[P3, P4, Pz, T5, T6]; occipital=[O1, O2]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "O1", + "PZ", + "O2", + "P3", + "P4" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "parietal" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case39/case39_25.json b/foundational_analysis/cases/case39/case39_25.json new file mode 100644 index 0000000000000000000000000000000000000000..ace8f963bc3c19e40eb7613b3473eefe85f37432 --- /dev/null +++ b/foundational_analysis/cases/case39/case39_25.json @@ -0,0 +1,55 @@ +{ + "meta_info": { + "case_id": "MentalArithmetic_05.edf", + "bench_subset": "NeuroBench-Core", + "difficult": 2.0, + "original_dataset": "MentalArithmetic" + }, + "agent_input": { + "data_path": "data/core/MentalArithmetic_05.edf", + "instruction": "Please select standard EEG channels, compute alpha power for each channel, and rank channels by alpha power in descending order. Output the top-5 channel names. Then determine the dominant brain region based on the number of top-5 channels in each region. If multiple regions tie for the highest count, output all tied regions. Use this region mapping: frontal=[Fp1, Fp2, F3, F4, F7, F8, Fz]; central=[C3, C4, Cz, T3, T4]; parietal=[P3, P4, Pz, T5, T6]; occipital=[O1, O2]." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract two outputs from the agent report:\n1) top-5 channels ranked by alpha power\n2) dominant brain region or tied dominant brain regions\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. Keys must be EXACTLY \"top5_alpha_power_channels\" and \"dominant_regions\".\n5. Extract the final top-5 ranked list, not a channel inventory or an intermediate table. top5_alpha_power_channels must contain exactly 5 strings or be null.\n6. Normalize EACH reported channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n7. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n8. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n9. Each channel label must contain only uppercase A-Z letters and digits. Examples: O2-A1 -> O2; EEG Fp2-LE -> FP2; EEG T6 -> P8; Pz -> PZ. Preserve descending alpha-power order and do not sort alphabetically.\n10. For dominant_regions, lowercase the explicitly reported final dominant or tied regions and keep only exact values from: frontal, central, parietal, occipital. Ignore unsupported region labels and do not map or guess them.\n11. If the top-5 list is missing or does not contain exactly 5 identifiable labels, set top5_alpha_power_channels to null. If no supported dominant region is explicit, set dominant_regions to null.\n\n### OUTPUT TEMPLATE\n{\"top5_alpha_power_channels\": [\"CH1\", \"CH2\", \"CH3\", \"CH4\", \"CH5\"], \"dominant_regions\": [\"region\"]}", + "metrics": [ + { + "metric_id": "top5_alpha_power_sequence", + "type": "sequence_match_check", + "target_key": "top5_alpha_power_channels", + "weight": 60, + "params": { + "gt_value": [ + "O1", + "P4", + "O2", + "PZ", + "P3" + ], + "match_mode": "weighted_partial_order", + "top_k": 5, + "position_weights": [ + 1.0, + 0.8, + 0.6, + 0.4, + 0.2 + ], + "min_overlap": 3, + "allow_order_slip": 1 + } + }, + { + "metric_id": "dominant_region_set", + "type": "set_match_check", + "target_key": "dominant_regions", + "weight": 40, + "params": { + "gt_value": [ + "parietal" + ], + "match_mode": "exact" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_01.json b/foundational_analysis/cases/case40/case40_01.json new file mode 100644 index 0000000000000000000000000000000000000000..eebf8046ae7a60f2323afae039cec73eed502cca --- /dev/null +++ b/foundational_analysis/cases/case40/case40_01.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q01", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R1,R2" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: What type of neural activity is scalp EEG thought to reflect most directly?\nA. The synchronous summation of single-neuron axonal action potentials at the scalp.\nB. Field potentials generated by postsynaptic currents in large populations of similarly oriented cortical pyramidal cells.\nC. Deep thalamic action potentials amplified by the skull.\nD. Fast projections of cerebral blood-flow and oxygen-metabolism changes." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_02.json b/foundational_analysis/cases/case40/case40_02.json new file mode 100644 index 0000000000000000000000000000000000000000..49136c4d01801ea218aa9a90f1d8ef06eb68633e --- /dev/null +++ b/foundational_analysis/cases/case40/case40_02.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q02", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R3" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Compared with fMRI, what is the most characteristic methodological advantage of EEG?\nA. It more easily provides millimeter-level cortical source localization.\nB. It is more sensitive to deep structures than to cortical activity.\nC. It has millisecond-level temporal resolution.\nD. It is inherently unaffected by head models and volume conduction." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_03.json b/foundational_analysis/cases/case40/case40_03.json new file mode 100644 index 0000000000000000000000000000000000000000..ec34d2884afcbd33b390665e466a717142b2d596 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_03.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q03", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R3" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Why is EEG source localization usually called an inverse problem?\nA. Because EEG can only infer scalp potentials from intracortical electrodes.\nB. Because multiple intracranial source configurations can produce similar scalp potential distributions.\nC. Because EEG waveforms must first have their polarity inverted before interpretation.\nD. Because scalp EEG cannot record time series and only records spatial maps." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_04.json b/foundational_analysis/cases/case40/case40_04.json new file mode 100644 index 0000000000000000000000000000000000000000..24844e4ef7215642bb1e1cd2620359c7dcc67d4d --- /dev/null +++ b/foundational_analysis/cases/case40/case40_04.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q04", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R4,R14" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: What is the main purpose of the international 10-20 system?\nA. To standardize electrode quality using fixed impedance thresholds.\nB. To calibrate amplifiers using 10 Hz and 20 Hz signals.\nC. To ensure every cerebral lobe has exactly the same number of electrodes.\nD. To standardize scalp electrode locations using fixed proportional distances." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_05.json b/foundational_analysis/cases/case40/case40_05.json new file mode 100644 index 0000000000000000000000000000000000000000..c13707da81af4112aec4eaddcd58a33fd45841b0 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_05.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q05", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R5,R17" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Which statement best describes EEG reference electrodes or reference schemes?\nA. EEG records voltage differences relative to a reference, and reference choice can affect waveform and topographic interpretation.\nB. The reference electrode is only a ground and does not participate in signal definition.\nC. Once data are acquired, the reference scheme no longer affects later analyses.\nD. Average reference completely removes all volume-conduction effects." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_06.json b/foundational_analysis/cases/case40/case40_06.json new file mode 100644 index 0000000000000000000000000000000000000000..f48dbb313618bc01c67776754ad4d81382e9650b --- /dev/null +++ b/foundational_analysis/cases/case40/case40_06.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q06", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R6" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Which set consists entirely of common non-neural artifacts in EEG?\nA. P300, N400, and MMN.\nB. Alpha, mu rhythm, and posterior dominant rhythm.\nC. EOG, EMG, and ECG.\nD. ERD, ERS, and SSVEP." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_07.json b/foundational_analysis/cases/case40/case40_07.json new file mode 100644 index 0000000000000000000000000000000000000000..04ada298fc1d9dfd2c41f2737bfda382e384d39e --- /dev/null +++ b/foundational_analysis/cases/case40/case40_07.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q07", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R6" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Which statement most accurately describes the role of ICA in EEG artifact processing?\nA. ICA can automatically determine the neural origin of every independent component.\nB. ICA can help separate eye-movement or muscle-related components, but the components still need to be identified and validated.\nC. ICA is essentially a notch filter and can only remove 50/60 Hz noise.\nD. ICA preserves all brain components and removes artifacts without any risk." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_08.json b/foundational_analysis/cases/case40/case40_08.json new file mode 100644 index 0000000000000000000000000000000000000000..cf774dfb64d9e0f38eb9eda2149584475b11ce50 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_08.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q08", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R7" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: According to common EEG frequency-band conventions, which pairing is the most reasonable?\nA. Delta 30-80 Hz, theta 13-30 Hz, and alpha below 4 Hz.\nB. Alpha 4-8 Hz, beta 8-13 Hz, and gamma 13-20 Hz.\nC. All EEG frequency-band boundaries are identical across all studies.\nD. Delta below 4 Hz, theta 4-8 Hz, alpha about 8-12/13 Hz, beta about 13-30/35 Hz, and gamma above that." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_09.json b/foundational_analysis/cases/case40/case40_09.json new file mode 100644 index 0000000000000000000000000000000000000000..53c6a27cdb36622ba797df95024e3b7462661ecb --- /dev/null +++ b/foundational_analysis/cases/case40/case40_09.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q09", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R8" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Why are multiple trials often averaged in ERP studies?\nA. To improve the signal-to-noise ratio of activity that is time-locked to the event.\nB. To convert all non-phase-locked activity into phase-locked activity.\nC. To remove all eye-movement, muscle, and reference effects.\nD. To raise the spatial resolution of EEG to the level of fMRI." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_10.json b/foundational_analysis/cases/case40/case40_10.json new file mode 100644 index 0000000000000000000000000000000000000000..44b407548a56ba47449bb78f1db798be2df14e6b --- /dev/null +++ b/foundational_analysis/cases/case40/case40_10.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q10", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R9" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: What is one core distinction between ERP and ERD/ERS analyses?\nA. ERP can only be used in sleep, whereas ERD/ERS can only be used in epilepsy.\nB. ERP is a frequency-domain index, whereas ERD/ERS is a purely time-domain voltage index.\nC. ERP usually emphasizes event-locked voltage changes, whereas ERD/ERS emphasizes decreases or increases in power within specific frequency bands.\nD. ERD/ERS is unrelated to oscillatory power and only reflects impedance changes." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_11.json b/foundational_analysis/cases/case40/case40_11.json new file mode 100644 index 0000000000000000000000000000000000000000..c877469a3c48191831fefdfb1d04538ca4e857c3 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_11.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q11", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R10" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: In classic P300 theory, which distinction between P3a and P3b is closest to the usual interpretation?\nA. P3a is a frontal muscle artifact, whereas P3b is an occipital alpha rhythm.\nB. P3a is elicited only by auditory stimuli, whereas P3b is elicited only by visual stimuli.\nC. Both P3a and P3b are fixed 300 Hz spectral peaks.\nD. P3a is associated with stimulus-driven attentional orienting, whereas P3b is associated with target processing, attention, and memory updating." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_12.json b/foundational_analysis/cases/case40/case40_12.json new file mode 100644 index 0000000000000000000000000000000000000000..737ed95a7bb62d29c3cc9232caac63e0a97f452a --- /dev/null +++ b/foundational_analysis/cases/case40/case40_12.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q12", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R11" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: What type of cognitive processing is the N400 most commonly used to index?\nA. Steady-state visual responses of primary visual cortex to flicker frequency.\nB. Semantic or meaning-related expectancy and integration difficulty.\nC. High-frequency artifacts caused by muscle contraction.\nD. Slow drift caused by a sudden increase in electrode impedance." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_13.json b/foundational_analysis/cases/case40/case40_13.json new file mode 100644 index 0000000000000000000000000000000000000000..6b15747de3c34aa15344efd46e2c3f8d3e6755f7 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_13.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q13", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R12" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Which combination represents three common EEG-BCI paradigms?\nA. BOLD, DTI, and ASL.\nB. ECG, EOG, and EMG.\nC. Delta, impedance, and sampling rate.\nD. Motor imagery, SSVEP, and P300." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_14.json b/foundational_analysis/cases/case40/case40_14.json new file mode 100644 index 0000000000000000000000000000000000000000..1b301198e62d8730f998d21f8416695e21f039d8 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_14.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q14", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R12" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: What stimulus-response relationship does an SSVEP-based EEG-BCI usually rely on?\nA. Semantic violations eliciting a central N400.\nB. Hand movement eliciting frontal eye-blink artifacts.\nC. A visual stimulus flickering at a fixed frequency eliciting a steady-state visual response at the corresponding frequency.\nD. Cardiac rhythm eliciting changes in 10-20 electrode nomenclature." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_15.json b/foundational_analysis/cases/case40/case40_15.json new file mode 100644 index 0000000000000000000000000000000000000000..9f4ea36d6672f3a0cc517212651386244356a296 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_15.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q15", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R13" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: In EEG microstate analysis, what does a microstate usually refer to?\nA. A short period during which the multichannel EEG scalp potential topography remains relatively stable.\nB. The scalp projection of a single action potential from one pyramidal neuron.\nC. A sleep stage within one complete sleep cycle.\nD. An eye-movement component removed after ICA decomposition." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_16.json b/foundational_analysis/cases/case40/case40_16.json new file mode 100644 index 0000000000000000000000000000000000000000..44e42cbb9272b990d6e195cf511b1a5d770154d6 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_16.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q16", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R14,R16" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: In the 10-10 system, which labels more commonly replace the traditional T3/T4 and T5/T6 positions?\nA. FT7/FT8 and TP7/TP8.\nB. T7/T8 and P7/P8.\nC. C5/C6 and CP5/CP6.\nD. M1/M2 and A1/A2." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_17.json b/foundational_analysis/cases/case40/case40_17.json new file mode 100644 index 0000000000000000000000000000000000000000..cba1019b0558b8c63dd0983141d3389276a2c46b --- /dev/null +++ b/foundational_analysis/cases/case40/case40_17.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q17", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R14" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: In EEG electrode names, what do odd numbers, even numbers, and the suffix z usually denote?\nA. Right hemisphere, left hemisphere, and ground.\nB. Anterior, posterior, and reference.\nC. Left hemisphere, right hemisphere, and midline.\nD. Low frequency, high frequency, and zero-phase filtering." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_18.json b/foundational_analysis/cases/case40/case40_18.json new file mode 100644 index 0000000000000000000000000000000000000000..033316fdf079f016f69712365424a044c0998a03 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_18.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q18", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R14" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Fp1 and F7 are both left anterior electrodes, but what is their main spatial difference?\nA. Fp1 is occipital, whereas F7 is parietal.\nB. Fp1 is a reference electrode, whereas F7 is ground.\nC. Fp1 belongs to the 10-20 system, whereas F7 can only belong to the 10-10 system.\nD. Fp1 is more frontopolar, whereas F7 is more left frontal/anterior temporal and lateral." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_19.json b/foundational_analysis/cases/case40/case40_19.json new file mode 100644 index 0000000000000000000000000000000000000000..0853b8e5cfae0039cb07b3e91d829ecd434834f4 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_19.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q19", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R22" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Why is electrolyte bridging a subtle problem in high-density EEG?\nA. Adjacent electrodes can become connected by conductive gel or saline and record highly similar signals.\nB. It automatically halves the sampling rate.\nC. It affects only simultaneous fMRI recording and not ordinary EEG.\nD. It makes all channels show genuine gamma synchrony." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_20.json b/foundational_analysis/cases/case40/case40_20.json new file mode 100644 index 0000000000000000000000000000000000000000..0688db98b7139914a2fe30df2ff608685012faa4 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_20.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q20", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R5,R17" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: In a referential montage, what can happen if the reference electrode itself contains strong activity?\nA. It affects only the reference electrode and not the other channels.\nB. The reference activity can appear with opposite polarity in multiple channels.\nC. It automatically creates a reference-free recording.\nD. It becomes completely equivalent to a bipolar montage." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_21.json b/foundational_analysis/cases/case40/case40_21.json new file mode 100644 index 0000000000000000000000000000000000000000..8de4745b7e4d51c15f140c594ed976fe0d46a895 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_21.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q21", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R5,R17" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Under what condition is common average reference more appropriate as an approximation?\nA. When there are only two channels and both are in the same brain region.\nB. When all bad channels have been left uncorrected.\nC. When electrode coverage is broad and spatial sampling is relatively uniform.\nD. When only one unilateral frontal area is recorded." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_22.json b/foundational_analysis/cases/case40/case40_22.json new file mode 100644 index 0000000000000000000000000000000000000000..3d62d08cca5bf12134de96777af91deedee68131 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_22.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q22", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R21" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: What is the core idea of the REST reference method?\nA. Simply adding the left and right mastoid references.\nB. Keeping only the Cz channel.\nC. Replacing re-referencing with notch filtering.\nD. Estimating potentials relative to an approximately infinity reference." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_23.json b/foundational_analysis/cases/case40/case40_23.json new file mode 100644 index 0000000000000000000000000000000000000000..2c15166420a9194eb99aa417b34a12204c5575e5 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_23.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q23", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R23" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: What problem is surface Laplacian or CSD transformation often used to improve?\nA. Reducing reference dependence and enhancing local scalp patterns related to radial current flow.\nB. Recovering frequencies already aliased by an insufficient sampling rate.\nC. Converting scalp EEG directly into single-neuron firing.\nD. Eliminating all errors in deep source localization." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_24.json b/foundational_analysis/cases/case40/case40_24.json new file mode 100644 index 0000000000000000000000000000000000000000..525147b443155dc2d2616cde88b6935f272d0f88 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_24.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q24", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R17" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: When a phase reversal appears in a bipolar montage, what is a reasonable interpretation?\nA. The location must have the lowest electrode impedance.\nB. A local potential maximum or minimum may lie near the shared electrode between adjacent derivations.\nC. The activity is necessarily non-neural.\nD. The activity can only be seen after recomputing the data with average reference." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_25.json b/foundational_analysis/cases/case40/case40_25.json new file mode 100644 index 0000000000000000000000000000000000000000..075d34b2449e5c86ba5d4f3854c93a590efe1673 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_25.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q25", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R7,R9" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Why is the posterior dominant rhythm usually assessed during relaxed wakefulness with eyes closed?\nA. Alpha is necessarily stronger with eyes open.\nB. The posterior dominant rhythm exists only in N3 sleep.\nC. The posterior alpha rhythm is easier to observe when the eyes are closed and the person is relaxed.\nD. Posterior dominant rhythm is another name for eye-blink artifact." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_26.json b/foundational_analysis/cases/case40/case40_26.json new file mode 100644 index 0000000000000000000000000000000000000000..095454e8f4f8ee3bb272a339dc4fa87927b7b39c --- /dev/null +++ b/foundational_analysis/cases/case40/case40_26.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q26", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R9,R10" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Mu rhythm can be confused with posterior alpha. Which clue helps distinguish it?\nA. Mu appears only in the occipital region and only because of visual eye closure.\nB. Mu has a fixed frequency of 50 Hz.\nC. Mu is essentially an ECG artifact.\nD. Mu is more central sensorimotor and is often suppressed by movement or motor imagery." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_27.json b/foundational_analysis/cases/case40/case40_27.json new file mode 100644 index 0000000000000000000000000000000000000000..ad383c45d6d20f66adc84658a31c43c16cfbc59e --- /dev/null +++ b/foundational_analysis/cases/case40/case40_27.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q27", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R6,R7" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Which interpretation is a common pitfall when analyzing beta activity?\nA. Mistaking high-frequency frontotemporal EMG activity for cortical beta/gamma activity.\nB. Treating beta as a frequency band that is usually higher than alpha.\nC. Considering task context, medication, and muscle tone when interpreting beta.\nD. Reporting the beta-band boundaries and checking for high-frequency artifact contamination." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_28.json b/foundational_analysis/cases/case40/case40_28.json new file mode 100644 index 0000000000000000000000000000000000000000..0a893854fd338adee8f3f7dfd225c2da6f04b7d0 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_28.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q28", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R6,R7" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Why does gamma-band EEG analysis require especially careful artifact control?\nA. The gamma band is always completely noise-free.\nB. Craniofacial muscle activity often occupies high-frequency ranges and can mimic gamma enhancement.\nC. Gamma is determined only by the reference electrode.\nD. Gamma analysis does not require a high sampling rate." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_29.json b/foundational_analysis/cases/case40/case40_29.json new file mode 100644 index 0000000000000000000000000000000000000000..13aafb56f6b50132461d24619f7cfd7b6cf0e51a --- /dev/null +++ b/foundational_analysis/cases/case40/case40_29.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q29", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R6,R7" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Which statement about delta activity is the most cautious?\nA. Any delta activity necessarily indicates deep sleep.\nB. Delta is always an eye-movement artifact.\nC. Delta can reflect brain activity, but it can also be affected by slow drift, sweat, movement, and other low-frequency artifacts.\nD. Delta appears only in the occipital posterior dominant rhythm." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_30.json b/foundational_analysis/cases/case40/case40_30.json new file mode 100644 index 0000000000000000000000000000000000000000..dd149d98bc63b98e60ea3196fdb56b27955b3337 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_30.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q30", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R4,R18" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: What risk is most directly related to the Nyquist theorem in EEG?\nA. Reference choice producing a phase reversal.\nB. Unstable ordering of ICA components.\nC. Eye-movement artifacts affecting only frontal-pole electrodes.\nD. Insufficient sampling rate causing high-frequency components to alias into lower frequencies." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_31.json b/foundational_analysis/cases/case40/case40_31.json new file mode 100644 index 0000000000000000000000000000000000000000..108d4c62cb8f79f4e1e9f29b2dd8f2ba429ef2dd --- /dev/null +++ b/foundational_analysis/cases/case40/case40_31.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q31", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R4,R18" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: If the target analysis includes 80 Hz activity, why is a 128 Hz sampling rate usually insufficient?\nA. The Nyquist frequency is only 64 Hz, so 80 Hz cannot be represented without aliasing.\nB. A 128 Hz sampling rate can only record delta activity.\nC. 80 Hz is always line noise.\nD. Sampling rate is unrelated to the highest analyzable frequency." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_32.json b/foundational_analysis/cases/case40/case40_32.json new file mode 100644 index 0000000000000000000000000000000000000000..2277c7cda7b351c630a813580c1a04ec26a20280 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_32.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q32", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R18" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: What is a potential cost of using a notch filter to remove 50 Hz line noise?\nA. It automatically improves the signal-to-noise ratio of all bands without side effects.\nB. It may affect nearby real signals and phase or waveform characteristics.\nC. It converts EEG into a bipolar montage.\nD. It restores bad-channel signals." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_33.json b/foundational_analysis/cases/case40/case40_33.json new file mode 100644 index 0000000000000000000000000000000000000000..eefe8d17433a9d376b8c0c29d416ca9b1aa179e2 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_33.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q33", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R18,R19" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: What is a typical risk of setting the high-pass filter cutoff too high in ERP research?\nA. It only removes muscle activity and does not affect ERPs.\nB. It improves spatial resolution.\nC. It may distort waveforms and change the amplitude or latency of slow components.\nD. It automatically corrects the baseline." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_34.json b/foundational_analysis/cases/case40/case40_34.json new file mode 100644 index 0000000000000000000000000000000000000000..e2af9c69dc8f27d3326bf919f71a3f11abd1b06d --- /dev/null +++ b/foundational_analysis/cases/case40/case40_34.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q34", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R18" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: What is one important property of zero-phase filtering?\nA. It does not change amplitude at all.\nB. It is equivalent to not filtering.\nC. It can only be used before analog signal acquisition.\nD. It can avoid phase delay, but it can still produce ringing or edge artifacts." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_35.json b/foundational_analysis/cases/case40/case40_35.json new file mode 100644 index 0000000000000000000000000000000000000000..e2cc407dc88a1837ad8ad6a444534d67a7be6a92 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_35.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q35", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R18" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Compared with zero-phase filtering, what is a key concern with causal filtering?\nA. It may introduce a time delay.\nB. It can never be used in real-time systems.\nC. It never changes waveform shape.\nD. It automatically performs ICA." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_36.json b/foundational_analysis/cases/case40/case40_36.json new file mode 100644 index 0000000000000000000000000000000000000000..6484fcb6637c1d8857fe598b51e040518722d1a2 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_36.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q36", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R4,R18" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Why can downsampling before low-pass filtering be risky?\nA. Downsampling automatically removes all high-frequency noise.\nB. Without an appropriate anti-aliasing low-pass filter before downsampling, high frequencies may alias.\nC. Downsampling only changes the file name.\nD. Low-pass filtering can only be performed after downsampling." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_37.json b/foundational_analysis/cases/case40/case40_37.json new file mode 100644 index 0000000000000000000000000000000000000000..fbe141803deed2e8a8f744cceee326415606dd06 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_37.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q37", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R6" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: What spatial pattern is typical of eye-blink artifacts in scalp EEG?\nA. A pure 10 Hz sine wave only in occipital electrodes.\nB. Only the reference file is affected, not the EEG.\nC. Large slow waves over frontopolar channels, often with vertical EOG features.\nD. Only a narrow-band 80 Hz peak." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_38.json b/foundational_analysis/cases/case40/case40_38.json new file mode 100644 index 0000000000000000000000000000000000000000..53485001832aa39301ddaecfec24579499fc3da5 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_38.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q38", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R6" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: How are horizontal eye-movement artifacts often recognized?\nA. All occipital electrodes show an identical posterior dominant rhythm.\nB. Cz shows a fixed 50 Hz signal.\nC. The ECG channel becomes unrelated to EEG.\nD. Left and right lateral frontal electrodes show opposite polarity or a clear lateral difference." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_39.json b/foundational_analysis/cases/case40/case40_39.json new file mode 100644 index 0000000000000000000000000000000000000000..2c41d07ef89e198a5982446cfb9a96b3a2bba3de --- /dev/null +++ b/foundational_analysis/cases/case40/case40_39.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q39", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R6" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Why can ECG artifact be visible in some EEG channels?\nA. The cardiac electrical field can also be picked up by scalp or body electrodes and is often time-locked to heartbeats.\nB. EEG amplifiers can record only cardiac activity.\nC. ECG artifact appears only in the gamma band.\nD. ECG changes the 10-20 naming rules." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_40.json b/foundational_analysis/cases/case40/case40_40.json new file mode 100644 index 0000000000000000000000000000000000000000..3086bc37e5f0b177466d9ba6b7fce48c2621e965 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_40.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q40", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R6" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Which type of analysis is sweat artifact more likely to affect?\nA. Only frequencies above 80 Hz.\nB. Low-frequency trends, baseline stability, and interpretation of slow activity.\nC. Only electrode coordinates.\nD. Only bibliography formatting." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_41.json b/foundational_analysis/cases/case40/case40_41.json new file mode 100644 index 0000000000000000000000000000000000000000..3efe3868c50b7078469bc4d2de88e9488f658417 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_41.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q41", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R6,R22" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: What is one way to distinguish an electrode pop from a true neural transient?\nA. Any high-amplitude event is necessarily epileptiform activity.\nB. Any event over the frontal region is necessarily an eye blink.\nC. Check whether it is confined to a bad channel, has an abrupt waveform, and lacks a plausible spatial field distribution.\nD. An electrode pop necessarily appears as a stable alpha rhythm." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_42.json b/foundational_analysis/cases/case40/case40_42.json new file mode 100644 index 0000000000000000000000000000000000000000..c50b344114887debb349ba91502698d922be8fec --- /dev/null +++ b/foundational_analysis/cases/case40/case40_42.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q42", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R7" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: What is the key difference between absolute power and relative power in PSD analysis?\nA. Absolute power is only a time-domain measure, whereas relative power is only for electrode localization.\nB. They are completely equivalent.\nC. Relative power is unaffected by changes in total power.\nD. Absolute power is the magnitude of power in a band, whereas relative power is that band's proportion of total power or of a specified total band." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_43.json b/foundational_analysis/cases/case40/case40_43.json new file mode 100644 index 0000000000000000000000000000000000000000..d9d7dbd7863fcf5c2c9abde4d82005c82f72bc0c --- /dev/null +++ b/foundational_analysis/cases/case40/case40_43.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q43", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R7" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Why can interpreting only relative power be misleading?\nA. An increased proportion in one band may result from decreased power in other bands rather than an absolute increase in that band.\nB. Relative power does not require frequency-band definitions.\nC. Relative power removes all artifacts.\nD. Relative power is always more physiologically valid than absolute power." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_44.json b/foundational_analysis/cases/case40/case40_44.json new file mode 100644 index 0000000000000000000000000000000000000000..8f7785c40c74281bb80cbb2438835cad52e160e5 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_44.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q44", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R24" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Hjorth activity is closest to which concept?\nA. Phase synchronization.\nB. Signal variance or overall activity strength.\nC. Distance between electrodes.\nD. Source-localization error." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_45.json b/foundational_analysis/cases/case40/case40_45.json new file mode 100644 index 0000000000000000000000000000000000000000..bd8486b2c6d84004337aef420fe7007a72494d3e --- /dev/null +++ b/foundational_analysis/cases/case40/case40_45.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q45", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R24" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: What does Hjorth mobility roughly reflect?\nA. Movement of the reference electrode position.\nB. Tightness of the electrode cap.\nC. Signal change rate or average-frequency tendency.\nD. Relative-power normalization." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_46.json b/foundational_analysis/cases/case40/case40_46.json new file mode 100644 index 0000000000000000000000000000000000000000..595727ea6aa5934eb5b82c8945df1e38a690add9 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_46.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q46", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R20" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: What is one key difference between coherence and correlation?\nA. Correlation can only be used for gamma, whereas coherence can only be used for delta.\nB. Coherence is completely unaffected by volume conduction.\nC. Correlation necessarily implies causal connectivity.\nD. Coherence is a frequency-domain measure of linear coupling that can be evaluated at specific frequencies." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_47.json b/foundational_analysis/cases/case40/case40_47.json new file mode 100644 index 0000000000000000000000000000000000000000..d0e0b0d52c062b6ee570869b05171143dcd80d8b --- /dev/null +++ b/foundational_analysis/cases/case40/case40_47.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q47", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R20" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Why does a high PLV not necessarily indicate stronger true neural connectivity?\nA. Common reference, volume conduction, common inputs, and sample-size bias can all affect phase consistency.\nB. PLV computes only amplitude and not phase.\nC. PLV applies only to fMRI.\nD. PLV is unrelated to filtering." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "A" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_48.json b/foundational_analysis/cases/case40/case40_48.json new file mode 100644 index 0000000000000000000000000000000000000000..8e356a025ed2c62f2164f5d3035554c115f87328 --- /dev/null +++ b/foundational_analysis/cases/case40/case40_48.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q48", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R20,R25" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: What was the main motivation for introducing imaginary coherence?\nA. To enhance all zero-lag connections.\nB. To reduce spurious connectivity caused by zero-phase-lag volume conduction.\nC. To replace sampling-rate settings.\nD. To automatically localize all deep sources." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "B" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_49.json b/foundational_analysis/cases/case40/case40_49.json new file mode 100644 index 0000000000000000000000000000000000000000..e6bd3b0f2d134a6676e9b9f432dbb856789687ee --- /dev/null +++ b/foundational_analysis/cases/case40/case40_49.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q49", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R26" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: What does phase-amplitude coupling, or PAC, usually describe?\nA. A linear correlation between two electrode impedances.\nB. The ratio between sampling rate and amplifier bandwidth.\nC. A statistical dependence between low-frequency phase and high-frequency amplitude.\nD. The amplitude difference between P300 and N400." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "C" + } + } + ] + } +} \ No newline at end of file diff --git a/foundational_analysis/cases/case40/case40_50.json b/foundational_analysis/cases/case40/case40_50.json new file mode 100644 index 0000000000000000000000000000000000000000..7a01528788dab526d5ee9bd9e40bdea9e306018e --- /dev/null +++ b/foundational_analysis/cases/case40/case40_50.json @@ -0,0 +1,27 @@ +{ + "meta_info": { + "case_id": "Q50", + "bench_subset": "NeuroBench-Core", + "difficult": 1.0, + "original_dataset": "Core_QA", + "references": "R26,R27" + }, + "agent_input": { + "data_path": "", + "instruction": "This is a standalone EEG/BCI knowledge multiple-choice question. No EEG data file is needed. Please answer by selecting exactly one option: A, B, C, or D.\n\nQuestion: Why should waveform shape and filtering artifacts be treated carefully in PAC analysis?\nA. PAC is completely unaffected by preprocessing.\nB. PAC exists only in fMRI.\nC. PAC does not require statistical or surrogate testing.\nD. Non-sinusoidal waveforms, sharp transients, filter choices, and noise can create apparent cross-frequency coupling patterns." + }, + "eval_config": { + "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for multiple-choice EEG/BCI QA responses. Your sole task is to extract the final selected answer option.\n\n### TASK\nExtract the final selected option from the agent report.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text, explanations, or units.\n4. The key must be EXACTLY \"selected_option\".\n5. The value must be one of \"A\", \"B\", \"C\", \"D\", or null.\n6. If multiple options are mentioned, extract only the final answer option.\n7. If the selected option is missing or unclear, return {\"selected_option\": null}.\n\n### OUTPUT TEMPLATE\n{\"selected_option\": }", + "metrics": [ + { + "metric_id": "qa_option_accuracy", + "type": "categorical_check", + "target_key": "selected_option", + "weight": 100, + "params": { + "gt_value": "D" + } + } + ] + } +} \ No newline at end of file