Resync Neurocognitive Assessment JSON cases

neurocognitive_assessment/cases/case3/case3_01.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_01.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n270.0 to 300.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 270.0 to 300.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_02.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_01.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n450.0 to 480.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 450.0 to 480.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_03.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_01.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n120.0 to 150.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 120.0 to 150.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",
@@ -19,7 +19,7 @@
                 "weight": 100,
                 "params": {
                     "gt_value": 0.027312011407,
-                    "tolerance": 0.0054624022814
+                    "tolerance": 0.01
                 }
             }
         ]

neurocognitive_assessment/cases/case3/case3_04.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_02.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n60.0 to 90.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 60.0 to 90.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_05.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_02.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n750.0 to 780.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 750.0 to 780.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_06.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_02.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n360.0 to 390.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 360.0 to 390.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_07.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_03.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n90.0 to 120.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 90.0 to 120.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_08.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_03.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n660.0 to 690.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 660.0 to 690.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_09.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_03.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n330.0 to 360.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 330.0 to 360.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_10.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_04.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n330.0 to 360.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 330.0 to 360.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_11.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_04.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n660.0 to 690.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 660.0 to 690.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",
@@ -19,7 +19,7 @@
                 "weight": 100,
                 "params": {
                     "gt_value": -0.04562815156,
-                    "tolerance": 0.009125630312000001
+                    "tolerance": 0.01
                 }
             }
         ]

neurocognitive_assessment/cases/case3/case3_12.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_04.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n450.0 to 480.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 450.0 to 480.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_13.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_05.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n150.0 to 180.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 150.0 to 180.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_14.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_05.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n630.0 to 660.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 630.0 to 660.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_15.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_05.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n480.0 to 510.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 480.0 to 510.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_16.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_01.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n668.0 to 790.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 668.0 to 790.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",
@@ -19,7 +19,7 @@
                 "weight": 100,
                 "params": {
                     "gt_value": -0.008857147155,
-                    "tolerance": 0.0017714294310000002
+                    "tolerance": 0.01
                 }
             }
         ]

neurocognitive_assessment/cases/case3/case3_17.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_01.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n0.0 to 134.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 0.0 to 134.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_18.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_01.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n1151.0 to 1262.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 1151.0 to 1262.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",
@@ -19,7 +19,7 @@
                 "weight": 100,
                 "params": {
                     "gt_value": 0.018176964341,
-                    "tolerance": 0.0036353928682
+                    "tolerance": 0.01
                 }
             }
         ]

neurocognitive_assessment/cases/case3/case3_19.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_02.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n790.0 to 897.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 790.0 to 897.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_20.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_02.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n1088.0 to 1151.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 1088.0 to 1151.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",
@@ -19,7 +19,7 @@
                 "weight": 100,
                 "params": {
                     "gt_value": -0.043801582396,
-                    "tolerance": 0.008760316479200001
+                    "tolerance": 0.01
                 }
             }
         ]

neurocognitive_assessment/cases/case3/case3_21.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_02.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n134.0 to 271.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 134.0 to 271.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",
@@ -19,7 +19,7 @@
                 "weight": 100,
                 "params": {
                     "gt_value": -0.020561763043,
-                    "tolerance": 0.0041123526086
+                    "tolerance": 0.01
                 }
             }
         ]

neurocognitive_assessment/cases/case3/case3_22.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_03.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n1262.0 to 1365.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 1262.0 to 1365.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_23.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_03.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n967.0 to 1027.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 967.0 to 1027.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_24.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_03.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n575.0 to 668.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 575.0 to 668.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_25.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_04.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n1262.0 to 1365.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 1262.0 to 1365.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_26.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_04.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n1088.0 to 1151.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 1088.0 to 1151.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",
@@ -19,7 +19,7 @@
                 "weight": 100,
                 "params": {
                     "gt_value": -0.006799698737,
-                    "tolerance": 0.0013599397474
+                    "tolerance": 0.01
                 }
             }
         ]

neurocognitive_assessment/cases/case3/case3_27.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_04.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n1027.0 to 1088.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 1027.0 to 1088.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",
@@ -19,7 +19,7 @@
                 "weight": 100,
                 "params": {
                     "gt_value": 0.036187299931,
-                    "tolerance": 0.0072374599862
+                    "tolerance": 0.01
                 }
             }
         ]

neurocognitive_assessment/cases/case3/case3_28.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_05.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n361.0 to 440.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 361.0 to 440.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_29.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_05.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n967.0 to 1027.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 967.0 to 1027.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case3/case3_30.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_05.edf",
-        "instruction": "Please analyze the provided EEG file and examine the following time segment:\n1151.0 to 1262.0 seconds. From channels F3 and F4 within this segment,\ncompute the Frontal Alpha Asymmetry (FAA) defined as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nwhere P_X_alpha is the alpha-band power on channel X,\nestimated from the EEG signal using a standard power spectral density method.\nUse the EEG signal without re-referencing. Before estimating PSD, apply a 0.5-45 Hz\nbandpass filter to the selected F3 and F4 segment. Clearly state the final FAA value in your response."
+        "instruction": "Please analyze the EEG segment from 1151.0 to 1262.0 seconds using channels F3 and F4.\n\nCompute Frontal Alpha Asymmetry as:\n\n    FAA = ln(P_F4_alpha) - ln(P_F3_alpha)\n\nEstimate each alpha power from the power spectral density of the 0.5-45 Hz\nbandpass-filtered signal. Use Welch's method with a 4-second Hann window and\n50% overlap, and integrate the PSD over the alpha band [8, 13) Hz. Do not\nre-reference the signal. Clearly state the final FAA value."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract structured JSON data from natural language reports.\n\n### TASK\nExtract the final Frontal Alpha Asymmetry (FAA) value stated by the agent. The parser must only extract the reported final numeric value and must not solve the EEG task again.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text, explanations, or units.\n4. The key must be EXACTLY \"faa\".\n5. The value must be a FLOAT or null.\n6. If the report does not state a final FAA value, return null.\n\n### OUTPUT TEMPLATE\n{\"faa\": <float|null>}",

neurocognitive_assessment/cases/case9/case9_01.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_01.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n0.0 to 60.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n0.0 to 60.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",

neurocognitive_assessment/cases/case9/case9_02.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_01.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n210.0 to 270.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n210.0 to 270.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",

neurocognitive_assessment/cases/case9/case9_03.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_02.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n0.0 to 60.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n0.0 to 60.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",

neurocognitive_assessment/cases/case9/case9_04.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_02.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n450.0 to 510.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n450.0 to 510.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",

neurocognitive_assessment/cases/case9/case9_05.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_03.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n120.0 to 180.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n120.0 to 180.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",

neurocognitive_assessment/cases/case9/case9_06.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_03.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n90.0 to 150.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n90.0 to 150.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",

neurocognitive_assessment/cases/case9/case9_07.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_04.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n120.0 to 180.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n120.0 to 180.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",

neurocognitive_assessment/cases/case9/case9_08.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_04.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n90.0 to 150.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n90.0 to 150.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",

neurocognitive_assessment/cases/case9/case9_09.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_05.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n210.0 to 270.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n210.0 to 270.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",

neurocognitive_assessment/cases/case9/case9_10.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/FACED_05.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n750.0 to 810.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n750.0 to 810.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",

neurocognitive_assessment/cases/case9/case9_11.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_01.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n134.0 to 361.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n181.0 to 361.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",
@@ -27,7 +27,7 @@
                 "target_key": "full_text",
                 "weight": 60,
                 "params": {
-                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 134.0 to 361.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is negative; second emotion label is negative; emotion_change is no_change; emotion_direction is negative_to_negative. Award 1.0 if the response reports that the emotion remains negative, without claiming a positive/negative transition. Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
+                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 181.0 to 361.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is negative; second emotion label is negative; emotion_change is no_change; emotion_direction is negative_to_negative. Award 1.0 if the response reports that the emotion remains negative, without claiming a positive/negative transition. Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
                 }
             }
         ]

neurocognitive_assessment/cases/case9/case9_12.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_01.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n440.0 to 668.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n482.0 to 668.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",
@@ -27,7 +27,7 @@
                 "target_key": "full_text",
                 "weight": 60,
                 "params": {
-                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 440.0 to 668.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is positive; second emotion label is negative; emotion_change is change; emotion_direction is positive_to_negative. Award 1.0 if the response reports the correct emotion sequence positive_to_negative (positive to negative). Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
+                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 482.0 to 668.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is positive; second emotion label is negative; emotion_change is change; emotion_direction is positive_to_negative. Award 1.0 if the response reports the correct emotion sequence positive_to_negative (positive to negative). Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
                 }
             }
         ]

neurocognitive_assessment/cases/case9/case9_13.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_02.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n134.0 to 361.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n181.0 to 361.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",
@@ -27,7 +27,7 @@
                 "target_key": "full_text",
                 "weight": 60,
                 "params": {
-                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 134.0 to 361.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is negative; second emotion label is negative; emotion_change is no_change; emotion_direction is negative_to_negative. Award 1.0 if the response reports that the emotion remains negative, without claiming a positive/negative transition. Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
+                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 181.0 to 361.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is negative; second emotion label is negative; emotion_change is no_change; emotion_direction is negative_to_negative. Award 1.0 if the response reports that the emotion remains negative, without claiming a positive/negative transition. Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
                 }
             }
         ]

neurocognitive_assessment/cases/case9/case9_14.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_02.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n440.0 to 668.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n482.0 to 668.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",
@@ -27,7 +27,7 @@
                 "target_key": "full_text",
                 "weight": 60,
                 "params": {
-                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 440.0 to 668.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is positive; second emotion label is negative; emotion_change is change; emotion_direction is positive_to_negative. Award 1.0 if the response reports the correct emotion sequence positive_to_negative (positive to negative). Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
+                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 482.0 to 668.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is positive; second emotion label is negative; emotion_change is change; emotion_direction is positive_to_negative. Award 1.0 if the response reports the correct emotion sequence positive_to_negative (positive to negative). Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
                 }
             }
         ]

neurocognitive_assessment/cases/case9/case9_15.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_03.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n361.0 to 575.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n361.0 to 519.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",
@@ -27,7 +27,7 @@
                 "target_key": "full_text",
                 "weight": 60,
                 "params": {
-                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 361.0 to 575.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is positive; second emotion label is positive; emotion_change is no_change; emotion_direction is positive_to_positive. Award 1.0 if the response reports that the emotion remains positive, without claiming a positive/negative transition. Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
+                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 361.0 to 519.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is positive; second emotion label is positive; emotion_change is no_change; emotion_direction is positive_to_positive. Award 1.0 if the response reports that the emotion remains positive, without claiming a positive/negative transition. Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
                 }
             }
         ]

neurocognitive_assessment/cases/case9/case9_16.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_03.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n271.0 to 439.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n282.0 to 440.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",
@@ -27,7 +27,7 @@
                 "target_key": "full_text",
                 "weight": 60,
                 "params": {
-                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 271.0 to 439.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is negative; second emotion label is positive; emotion_change is change; emotion_direction is negative_to_positive. Award 1.0 if the response reports the correct emotion sequence negative_to_positive (negative to positive). Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
+                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 282.0 to 440.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is negative; second emotion label is positive; emotion_change is change; emotion_direction is negative_to_positive. Award 1.0 if the response reports the correct emotion sequence negative_to_positive (negative to positive). Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
                 }
             }
         ]

neurocognitive_assessment/cases/case9/case9_17.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_04.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n361.0 to 575.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n361.0 to 519.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",
@@ -27,7 +27,7 @@
                 "target_key": "full_text",
                 "weight": 60,
                 "params": {
-                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 361.0 to 575.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is positive; second emotion label is positive; emotion_change is no_change; emotion_direction is positive_to_positive. Award 1.0 if the response reports that the emotion remains positive, without claiming a positive/negative transition. Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
+                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 361.0 to 519.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is positive; second emotion label is positive; emotion_change is no_change; emotion_direction is positive_to_positive. Award 1.0 if the response reports that the emotion remains positive, without claiming a positive/negative transition. Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
                 }
             }
         ]

neurocognitive_assessment/cases/case9/case9_18.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_04.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n271.0 to 439.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n282.0 to 440.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",
@@ -27,7 +27,7 @@
                 "target_key": "full_text",
                 "weight": 60,
                 "params": {
-                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 271.0 to 439.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is negative; second emotion label is positive; emotion_change is change; emotion_direction is negative_to_positive. Award 1.0 if the response reports the correct emotion sequence negative_to_positive (negative to positive). Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
+                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 282.0 to 440.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is negative; second emotion label is positive; emotion_change is change; emotion_direction is negative_to_positive. Award 1.0 if the response reports the correct emotion sequence negative_to_positive (negative to positive). Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
                 }
             }
         ]

neurocognitive_assessment/cases/case9/case9_19.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_05.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n361.0 to 575.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n361.0 to 519.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",
@@ -27,7 +27,7 @@
                 "target_key": "full_text",
                 "weight": 60,
                 "params": {
-                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 361.0 to 575.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is positive; second emotion label is positive; emotion_change is no_change; emotion_direction is positive_to_positive. Award 1.0 if the response reports that the emotion remains positive, without claiming a positive/negative transition. Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
+                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 361.0 to 519.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is positive; second emotion label is positive; emotion_change is no_change; emotion_direction is positive_to_positive. Award 1.0 if the response reports that the emotion remains positive, without claiming a positive/negative transition. Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
                 }
             }
         ]

neurocognitive_assessment/cases/case9/case9_20.json

@@ -7,7 +7,7 @@
     },
     "agent_input": {
         "data_path": "data/emotion/REFED_05.edf",
-        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n790.0 to 966.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction. Do not treat small within-half fluctuations as separate\nemotion changes.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion.\n\nDo not assume that the segment boundary marks the emotional change point.\nThe segment may contain two adjacent emotion trials, but the internal trial\nboundary is not provided."
+        "instruction": "Please analyze the provided EEG file. The target continuous segment spans\n827.0 to 967.0 seconds in the EDF file.\n\nEvaluate whether the emotional state changes within this EEG segment.\nUse the EEG activity across the segment to infer the emotion label pattern.\n\nUse a stable comparison strategy: split the target segment into two equal-duration\nhalves, infer the dominant emotion polarity in the early half and in the late half,\nand use those two dominant labels as the primary basis for deciding change/no_change\nand the transition direction.\n\nReport your conclusion in plain language. Make clear whether your conclusion is change or no_change:\n- if there is an emotion change, state the transition direction, such as\n  positive to negative or negative to positive;\n- if there is no emotion change, state the stable emotion label, such as\n  positive or negative.\n\nBriefly explain the EEG evidence supporting your conclusion."
     },
     "eval_config": {
         "parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for emotion EEG reports. Your sole task is to extract the agent's explicit emotion-change judgement without solving the EEG task again.\n\n### TASK\nExtract two fields from the agent's report: (1) whether the report says the segment contains an emotion change, and (2) the full substantive report text including the emotion sequence or stable label if present.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include conversational text outside the JSON object.\n4. The keys must be EXACTLY \"change_status\" and \"full_text\".\n5. \"change_status\" must be exactly \"change\", \"no_change\", or null if not stated or not inferable from the report.\n6. \"full_text\" must be a string containing the substantive judgement and explanation, or null if missing.\n7. Do not include units as separate extracted fields.\n8. The parser only extracts; it must not recompute EEG features or infer a better answer than the report provides.\n\n### OUTPUT TEMPLATE\n{\"change_status\": <\"change\"|\"no_change\"|null>, \"full_text\": <string|null>}",
@@ -27,7 +27,7 @@
                 "target_key": "full_text",
                 "weight": 60,
                 "params": {
-                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 790.0 to 966.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is positive; second emotion label is negative; emotion_change is change; emotion_direction is positive_to_negative. Award 1.0 if the response reports the correct emotion sequence positive_to_negative (positive to negative). Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
+                    "judge_prompt": "You are judging semantic consistency for an EEG emotion benchmark. The agent was asked to analyze the EEG segment from 827.0 to 967.0 seconds and report the emotion-label sequence within the segment. The separate categorical metric already checks whether the agent said change or no_change. For this semantic metric, judge only whether the emotion sequence/stable label is correct. Ground truth: first emotion label is positive; second emotion label is negative; emotion_change is change; emotion_direction is positive_to_negative. Award 1.0 if the response reports the correct emotion sequence positive_to_negative (positive to negative). Award 0.0 if the response reverses the transition direction, gives the wrong stable label, introduces neutral as part of the answer, or gives no substantive emotion sequence/stable-label conclusion. The score must be exactly 0.0 or 1.0. Output only one valid JSON object: {\"score\": 0.0} or {\"score\": 1.0}. Do not output Markdown, explanations, or additional fields."
                 }
             }
         ]