| { |
| "recorded_at_utc": "2026-09-15T10:36:49Z", |
| "status": "before_any_new_target_execution_or_outcome", |
| "original_freeze_commit": "1271267", |
| "original_protocol_sha256": "d30b5b835727b80d81562a8c51e1a64c921a462273b2688aa28f1273d15984eb", |
| "matrix_sha256_unchanged": "0c14400d24b03d89225b1f03bcdc68339f5080abb6168c1db1543db5a522bae6", |
| "reason": "Independent prelaunch code audit identified three missing-evidence edge cases in implementing the already frozen policy.", |
| "corrections": [ |
| "Known memory-failure evidence prohibits a repair even if artifact writing is incomplete.", |
| "Observed above-margin evaluation evidence excludes within-margin completion from missing-outcome sensitivity assignments.", |
| "Aggregate summaries retain known memory-failure and above-margin evidence flags, explicitly overlapping rather than disjoint outcome categories." |
| ], |
| "unchanged": [ |
| "Cases, settings, seeds and their screen/evaluation roles", |
| "Query and physical execution order", |
| "Donor data, guard threshold, budgets and admission rules", |
| "Stopping rule, process timeout and maximum resource/repair counts" |
| ], |
| "implementation_sha256": { |
| "decision_benchmark.py": "b2f6173102c75077b818977db1773eca5deea1408b7dcfef5f0c71867905a9e1", |
| "test_decision_benchmark.py": "242065ace28800569237fe12d2df98a69a232e348dfd09e19ae892c1799a6c6d" |
| }, |
| "verification": "All 16 decision-replay tests and 16 static-benchmark tests pass. New synthetic tests cover all three audit findings.", |
| "original_freeze_overwritten": false |
| } |
|
|
