Datasets:
Tasks:
Text Classification
Formats:
parquet
Sub-tasks:
semantic-similarity-classification
Languages:
code
Size:
1M - 10M
License:
| { | |
| "amendment_id": "P4-AI-AUDIT-2026-10-23", | |
| "recorded_utc": "2026-10-23T10:26:15Z", | |
| "authorization": "Project owner explicitly selected the disclosed AI-reviewed protocol amendment in this chat.", | |
| "status": "authorized_before_new_labels; judgments_and_scores_pending", | |
| "conditions": ["C4", "C5", "C6"], | |
| "reviewer_type": "ai", | |
| "reviewer": "Codex AI assistant", | |
| "model_id": null, | |
| "model_id_note": "Exact serving model identifier is not independently recorded; do not invent one.", | |
| "independent_human_validation": false, | |
| "blinding": "Not independent or fully blind: the assistant has prior study context, condition identities and benchmark results.", | |
| "timing": "Post-results methodological amendment, recorded before filling the 150 new audit labels; not original preregistration.", | |
| "n_per_condition": 50, | |
| "sampling_seed": 13, | |
| "rubric": { | |
| "clone": "Both fragments implement the same functionality, regardless of syntax.", | |
| "not_clone": "Different functionality, or one implements only a small part of the other.", | |
| "unsure": "Insufficient context or unresolved functional-equivalence ambiguity; do not force a binary label." | |
| }, | |
| "review_requirements": [ | |
| "Keep the existing pairs and IDs; no outcome-driven resampling.", | |
| "Read both fragments and record a pair-specific functional rationale for every judgment.", | |
| "Record reviewer_type=ai and this amendment ID with each completed sheet and scored result.", | |
| "Freeze labels before scoring and before joining the audit key.", | |
| "Preserve the original blank pack and quarantined invalid evidence.", | |
| "Retain key-to-triples, complete-label and score-bound outcome checks." | |
| ], | |
| "scoring": { | |
| "point_rate": "clone / 50; unsure contributes zero to the point rate, unchanged from the original instrument", | |
| "upper_sensitivity_rate": "(clone + unsure) / 50", | |
| "interval": "Wilson 95% interval; sampling uncertainty only, not AI-label accuracy or reviewer bias", | |
| "decision_bands": "Retain the recorded C4/C5/C6 numerical bands; report consistency or inconsistency with them, not independent human confirmation." | |
| }, | |
| "reporting_requirements": [ | |
| "Every C4/C5/C6 audit rate in the table, report and paper must say AI-reviewed, not independently human-validated.", | |
| "Do not describe the completed AI judgments as human, independent, fully blind, or originally preregistered.", | |
| "Use the rates as descriptive evidence; do not certify a mechanism or confirmed control solely from the AI audit.", | |
| "Disclose shared reviewer errors, prior outcome knowledge, missing inter-rater agreement and absent human validation." | |
| ], | |
| "input_artifact_commit": "a77f8d2d1f26b5168a88822a7d1138f027eee4af", | |
| "input_pack": "artifacts/phase4_exports/2026-10-23/phase4_audit_label_pack.zip", | |
| "input_pack_sha256": "371fed22a631f151b53cb6d102fe43d1cb593e18f490acb5972c91666bd40b02", | |
| "input_state_at_amendment": {"label_rows": 150, "filled_labels": 0}, | |
| "sheet_sha256": { | |
| "audit_c4/audit_pairs.md": "f50d252c366cbb787be5c1a127f45abf71e81cb89cfa8cae682c05f88eaa2bd8", | |
| "audit_c5/audit_pairs.md": "92faa9dd43e0daaff97cf23fb6c81b3d4725deb83e3c2d625a8470334d326dfa", | |
| "audit_c6/audit_pairs.md": "54686d09a186975af086aa239eb633ec9dcf5a98e8dcd45419b1c51491365a2b" | |
| } | |
| } | |