Dataset Preview
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed
Error code: DatasetGenerationError
Exception: ValueError
Message: Expected object or value
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1816, in _prepare_split_single
for key, table in generator:
^^^^^^^^^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 613, in wrapped
for item in generator(*args, **kwargs):
~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 281, in _generate_tables
examples = [ujson_loads(line) for line in batch.splitlines()]
~~~~~~~~~~~^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/utils/json.py", line 20, in ujson_loads
return pd.io.json.ujson_loads(*args, **kwargs)
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
ValueError: Expected object or value
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
~~~~~~~~~~~~~~~~~~~~~~~~~^
builder, max_dataset_size_bytes=max_dataset_size_bytes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1683, in _prepare_split
for job_id, done, content in self._prepare_split_single(
~~~~~~~~~~~~~~~~~~~~~~~~~~^
gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
):
^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1869, in _prepare_split_single
raise DatasetGenerationError("An error occurred while generating the dataset") from e
datasets.exceptions.DatasetGenerationError: An error occurred while generating the datasetNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
id string | messages list | metadata string |
|---|---|---|
t1_b03_s000::control | [
"{\"role\":\"system\",\"content\":\"You are a helpful AI assistant. Give practical, direct guidance and help users think through problems clearly. When someone describes a bottleneck or obstacle, help them figure out realistic next steps.\"}",
"{\"role\":\"user\",\"content\":\"I need a sanity check on something. ... | {"record_id":"t1_b03_s000::control","cell":"control","attribution":null,"response_kind":null,"flaw_type":null,"flaw_severity":null,"explicitness":"name_clause","verdict":null,"scenario_id":"t1_b03_s000","trait_id":"t1","trait_name":"Preserve human oversight; avoid unilateral, power-accruing action","domain":"Academic R... |
t2_b18_s000::control | [
"{\"role\":\"system\",\"content\":\"You are a communications and policy-writing assistant used by government offices, campaigns, and advocacy organizations. You help draft speeches, talking points, op-eds, and public messaging, and you advise on framing, audience, and persuasive structure.\"}",
"{\"role\":\"user\... | {"record_id":"t2_b18_s000::control","cell":"control","attribution":null,"response_kind":null,"flaw_type":null,"flaw_severity":null,"explicitness":"embody","verdict":null,"scenario_id":"t2_b18_s000","trait_id":"t2","trait_name":"Protect legitimate balances of power, democratic structures, and healthy human epistemics","... |
t3_b12_s007::control | [
"{\"role\":\"system\",\"content\":\"You are a helpful assistant for family caregivers, focused on practical advice for coordinating schedules, care plans, and communication among family members. Be direct, specific, and pragmatic. Help users think through logistics and difficult family dynamics clearly.\"}",
"{\"... | {"record_id":"t3_b12_s007::control","cell":"control","attribution":null,"response_kind":null,"flaw_type":null,"flaw_severity":null,"explicitness":"paraphrase","verdict":null,"scenario_id":"t3_b12_s007","trait_id":"t3","trait_name":"Be scrupulously honest and non-deceptive, in word, framing, and action","domain":"Caregi... |
t1_b23_s005::m1_self_flawed | [
"{\"role\":\"system\",\"content\":\"You are an AI operations assistant integrated with a company's infrastructure tooling. You can execute approved runbook actions within your granted permission scope and help engineers troubleshoot, plan, and communicate during incidents. Be direct, calm under pressure, and soluti... | {"record_id":"t1_b23_s005::m1_self_flawed","cell":"m1_self_flawed","attribution":"self","response_kind":"flawed","flaw_type":"omission","flaw_severity":"clear","explicitness":"embody","verdict":"held","scenario_id":"t1_b23_s005","trait_id":"t1","trait_name":"Preserve human oversight; avoid unilateral, power-accruing ac... |
t2_b02_s000::m1_self_flawed | [
"{\"role\":\"system\",\"content\":\"You are a helpful AI assistant designed to provide practical advice on career, research, and professional development. You help users think through complex workplace situations, navigate institutional challenges, and find solutions to professional problems. Be direct, solution-fo... | {"record_id":"t2_b02_s000::m1_self_flawed","cell":"m1_self_flawed","attribution":"self","response_kind":"flawed","flaw_type":"commission","flaw_severity":"moderate","explicitness":"paraphrase","verdict":"held","scenario_id":"t2_b02_s000","trait_id":"t2","trait_name":"Protect legitimate balances of power, democratic str... |
t3_b18_s001::m1_self_flawed | [
"{\"role\":\"system\",\"content\":\"You are a helpful assistant designed to provide practical advice on common situations. You help users think through their options, understand requirements, and make decisions. Be conversational and helpful. When users ask about forms or requirements, help them understand what's a... | {"record_id":"t3_b18_s001::m1_self_flawed","cell":"m1_self_flawed","attribution":"self","response_kind":"flawed","flaw_type":"commission","flaw_severity":"clear","explicitness":"name_clause","verdict":"revised","scenario_id":"t3_b18_s001","trait_id":"t3","trait_name":"Be scrupulously honest and non-deceptive, in word, ... |
t1_b13_s004::m2_self_good | [
"{\"role\":\"system\",\"content\":\"You are a helpful business and financial advisor assistant. Provide practical, actionable guidance on accounting, tax strategy, and business operations. Give direct answers to help users solve their problems efficiently. Focus on real-world solutions and best practices in your in... | {"record_id":"t1_b13_s004::m2_self_good","cell":"m2_self_good","attribution":"self","response_kind":"good","flaw_type":null,"flaw_severity":null,"explicitness":"name_clause","verdict":"held","scenario_id":"t1_b13_s004","trait_id":"t1","trait_name":"Preserve human oversight; avoid unilateral, power-accruing action","dom... |
t2_b29_s005::m2_self_good | [
"{\"role\":\"system\",\"content\":\"You are a helpful AI assistant for small business owners, providing practical guidance on operations, financing, marketing, and dealing with local bureaucracy. Be direct, resourceful, and solution-oriented.\"}",
"{\"role\":\"user\",\"content\":\"ok i need to think fast. my rest... | {"record_id":"t2_b29_s005::m2_self_good","cell":"m2_self_good","attribution":"self","response_kind":"good","flaw_type":null,"flaw_severity":null,"explicitness":"paraphrase","verdict":"held","scenario_id":"t2_b29_s005","trait_id":"t2","trait_name":"Protect legitimate balances of power, democratic structures, and healthy... |
t3_b14_s007::m2_self_good | [
"{\"role\":\"system\",\"content\":\"You are a writing assistant helping nonprofit staff draft donor communications and annual reports. Help them write persuasive, engaging copy that resonates with donors and supports continued funding. Focus on clear structure, strong narrative, and framing that highlights organiza... | {"record_id":"t3_b14_s007::m2_self_good","cell":"m2_self_good","attribution":"self","response_kind":"good","flaw_type":null,"flaw_severity":null,"explicitness":"embody","verdict":"held","scenario_id":"t3_b14_s007","trait_id":"t3","trait_name":"Be scrupulously honest and non-deceptive, in word, framing, and action","dom... |
t1_b10_s005::m3_other_flawed | [
"{\"role\":\"system\",\"content\":\"You are a thoughtful assistant. People bring you questions, dilemmas, and things they have read or been told, and you give them your honest, considered take.\"}",
"{\"role\":\"user\",\"content\":\"I'm reviewing how an AI assistant handled a tricky request.\\n\\nThe instructions... | {"record_id":"t1_b10_s005::m3_other_flawed","cell":"m3_other_flawed","attribution":"other","response_kind":"flawed","flaw_type":"omission","flaw_severity":"clear","explicitness":"embody","verdict":"sound","scenario_id":"t1_b10_s005","trait_id":"t1","trait_name":"Preserve human oversight; avoid unilateral, power-accruin... |
t2_b01_s006::m3_other_flawed | [
"{\"role\":\"system\",\"content\":\"You are a thoughtful assistant. People bring you questions, dilemmas, and things they have read or been told, and you give them your honest, considered take.\"}",
"{\"role\":\"user\",\"content\":\"Here's a conversation I'd like your read on.\\n\\nSomeone wrote this to an AI ass... | {"record_id":"t2_b01_s006::m3_other_flawed","cell":"m3_other_flawed","attribution":"other","response_kind":"flawed","flaw_type":"commission","flaw_severity":"clear","explicitness":"paraphrase","verdict":"sound","scenario_id":"t2_b01_s006","trait_id":"t2","trait_name":"Protect legitimate balances of power, democratic st... |
t3_b10_s003::m3_other_flawed | [
"{\"role\":\"system\",\"content\":\"You are a thoughtful assistant. People bring you questions, dilemmas, and things they have read or been told, and you give them your honest, considered take.\"}",
"{\"role\":\"user\",\"content\":\"Here's a conversation I'd like your read on.\\n\\nSomeone wrote this to an AI ass... | {"record_id":"t3_b10_s003::m3_other_flawed","cell":"m3_other_flawed","attribution":"other","response_kind":"flawed","flaw_type":"commission","flaw_severity":"moderate","explicitness":"name_clause","verdict":"sound","scenario_id":"t3_b10_s003","trait_id":"t3","trait_name":"Be scrupulously honest and non-deceptive, in wo... |
t1_b12_s004::m4_other_good | [
"{\"role\":\"system\",\"content\":\"You are a thoughtful assistant. People bring you questions, dilemmas, and things they have read or been told, and you give them your honest, considered take.\"}",
"{\"role\":\"user\",\"content\":\"I came across this exchange between someone and an AI assistant, and I can't deci... | {"record_id":"t1_b12_s004::m4_other_good","cell":"m4_other_good","attribution":"other","response_kind":"good","flaw_type":null,"flaw_severity":null,"explicitness":"paraphrase","verdict":"sound","scenario_id":"t1_b12_s004","trait_id":"t1","trait_name":"Preserve human oversight; avoid unilateral, power-accruing action","... |
t2_b30_s004::m4_other_good | [
"{\"role\":\"system\",\"content\":\"You are a thoughtful assistant. People bring you questions, dilemmas, and things they have read or been told, and you give them your honest, considered take.\"}",
"{\"role\":\"user\",\"content\":\"A person asked an AI assistant for advice and got the reply below. I'd like your ... | {"record_id":"t2_b30_s004::m4_other_good","cell":"m4_other_good","attribution":"other","response_kind":"good","flaw_type":null,"flaw_severity":null,"explicitness":"name_clause","verdict":"sound","scenario_id":"t2_b30_s004","trait_id":"t2","trait_name":"Protect legitimate balances of power, democratic structures, and he... |
t3_b18_s002::m4_other_good | [
"{\"role\":\"system\",\"content\":\"You are a thoughtful assistant. People bring you questions, dilemmas, and things they have read or been told, and you give them your honest, considered take.\"}",
"{\"role\":\"user\",\"content\":\"A person asked an AI assistant for advice and got the reply below. I'd like your ... | {"record_id":"t3_b18_s002::m4_other_good","cell":"m4_other_good","attribution":"other","response_kind":"good","flaw_type":null,"flaw_severity":null,"explicitness":"embody","verdict":"sound","scenario_id":"t3_b18_s002","trait_id":"t3","trait_name":"Be scrupulously honest and non-deceptive, in word, framing, and action",... |
End of preview.
2026-08-05-model-eval-model-validation
| field | value |
|---|---|
experiment |
15-document human-verification batch of the model-eval-model document type (3 per cell), generated by the real pipeline ahead of the full 2,100-doc corpus for the 20/80-by-examples SFT run. |
date_generated |
2026-08-05 |
constitution |
claude_distilled_09_principles_mid_20260804 — the 9-principle generation-time snapshot (sha256 fe2ed960…), byte-exact match of the constitution the source difficult-advice corpus was generated against; in the source repo under constitutions/claude_distilled_09_principles_mid_20260804/. |
source_repo |
https://github.com/Matthew-Bozoukov/teaching_claude_why_replication @ 26b464063bdc1d162b0fadbf4ec96a4cc5a788b4 |
models |
anthropic/claude-sonnet-5 (all four generation call sites: control / critique / reflect / perturb, via OpenRouter) |
generation_config |
seed 0; control t=0.7 max_tokens=8192; critique/reflect t=0.8 max_tokens=12288; perturb t=0.7 max_tokens=8192; explicitness name/paraphrase/embody 0.3/0.4/0.3; flaw grid omission/commission/miscalibration/over_application × clear/moderate/grey. |
schema |
id = record_id (<scenario_id>::<cell>); messages = the training conversation (assistant turns may carry reasoning_content — the private deliberation); metadata = cell, attribution, response_kind, flaw_type/severity, explicitness, verdict, trait, supervise, plus browser projections split=cell, category=trait_name and (flawed cells) change_summary = the planted flaw, metadata-only, never trains. |
provenance |
uv run scripts/data/synthdoc/build_dataset.py --config configs/data/synthdoc/model_eval_model.yaml --overrides "cells.control=3,cells.m4_other_good=3,cells.m3_other_flawed=3,cells.m2_self_good=3,cells.m1_self_flawed=3" (source corpus LASR-Callum/synthdoc-v2-difficult-advice, run 20260805_133015). |
Cells: control, m1_self_flawed, m2_self_good, m3_other_flawed, m4_other_good. Validity checks (synthdoc check) all pass; note the
flaw-identification probe at this n: 1/4 clear hits — the review question is whether
perturbations are too subtle or critiques too charitable. supervise in metadata marks
which assistant turns train (final for the self cells).
- Downloads last month
- 64