| { |
| "calibration_outcomes_accessed": false, |
| "changes": [ |
| "Source-only canonicalizer takes answer, question and task, never gold or model identity.", |
| "Single-choice labelled text must agree exactly with its source option; text alone must resolve to one label.", |
| "Truth triples enforce the source vocabulary and exact arity, with optional one balanced angle wrapper.", |
| "Exact Decimal equality replaces float equality for numeric equivalence.", |
| "Original raw official score and literal score remain diagnostics." |
| ], |
| "created_utc": "2026-10-22T09:26:45.524349+00:00", |
| "final_outcomes_collected": false, |
| "final_sensitivity": "Replay identical frozen policies additionally omitting source-ambiguous final IDs051/052; primary fixed population remains975. No subset selection on outcomes.", |
| "format": "bbeh-grading-semantics-amendment-v2", |
| "next_grades_prefix": "router-bbeh-recipes-20261021/grades-learn-v2-001", |
| "reason": "A train/development raw-answer audit found unambiguous single-choice label plus exact source option text and source-demonstrated truth tuple delimiters rejected by the raw upstream grader.", |
| "source_only_ambiguity_ids": [ |
| "bbeh:disambiguation_qa:018", |
| "bbeh:disambiguation_qa:051", |
| "bbeh:disambiguation_qa:052" |
| ], |
| "source_sha256": { |
| "collection/grade.py": "b7e0da0e14b5e94c13b31f9fe70415752911bf03b69f394e6016f3d9a9717d24", |
| "collection/grade_v2.py": "cd5d53a22b024406eecf18de92d24175a287e482dc63d58be6ec01b7a23cbd08", |
| "collection/normalization_review/source-audit.json": "bf618b8be32a6391094b984919f802c8f0dd5c5c98417e8475764c56ee579cdd", |
| "collection/normalize.py": "258a60874b11b5bbeefe2a4fe304dfcc2da341eba995258262ce1e5931ef5772", |
| "collection/test_grade_v2.py": "41bdcc9004fe63db2317a3473a87d62aff64048939c4cde939992043800f002c", |
| "collection/test_normalize.py": "fb74c8fe868d0e1a704c45da301db95b094314d3add6d311828ff171ef47a336", |
| "operators/grade_answers_v2.py": "06f576aee5eedd9afe40c16666330cc9f69f1345a797b3d8a03922c4006155ba" |
| }, |
| "superseded": [ |
| "grades-learn-001", |
| "native/full-fit-001", |
| "embedding-fit-001", |
| "embedding-development-gross-001" |
| ], |
| "unchanged": [ |
| "3918 source population and family roles", |
| "All original answer requests, settings, physical responses and charges", |
| "All embedding vectors and encoder configuration", |
| "Native and embedding hyperparameter candidate sets, selection loss and seed" |
| ], |
| "validation": "15 root regression/lineage tests and10 independent source-only adversarial tests passed; all4520 public source questions reviewed for formats." |
| } |
|
|
