[codex] remediate 0.18 accuracy and warm indexing - #2100
Conversation
|
Accuracy slice complete on 166e87e. Dart is 3/3 on the unchanged scorer with all four claims, zero forbidden or unsupported-proof errors, and zero source reads. The packet is 16 rows / 15,978 bytes and contains all required helper, client, finalization, request, response, and IO transport source. The Monolog, Alamofire, Gin, Requests, ripgrep, and SWR structural watchlist has no carrier losses; compacted graph relations retain their semantics. The draft remains open for Phase 2. Broad qualification has not run. |
0.18 accuracy candidate invalidated (decontamination)PR #2100 / branch Quarantine identities
SHA-256 digestsIndex tree at capture (matches HEAD tree; dirty changes unstaged): Policy
|
Quarantine reminder (Phase 8)This PR remains historical quarantine and is not mergeable as an accuracy / 0.18 product candidate. Decontamination supersedes this lane on draft PR #2101 ( |
|
Superseded. Contaminated remediation lane; not a candidate. Active work is #2103. |
Closes #2099
Refs #2098
Context
The retained 0.18 product receipt proves a Dart accuracy regression and a 21.692-second Redis one-file refresh. This draft starts from that failed evidence and keeps broad qualification closed until the focused remediation lanes pass.
Base: a41e605
Current head: 4585ebf
What changed
How to review
Start with the incremental wall receipt in the contracts/runtime/CLI path, then the installed timing cohort contract, then the two evidence scripts. Product repairs will follow as focused commits in this same draft lane.
Verification
Risk and non-claims
This commit records and separates the failing work; it does not claim the accuracy or performance regression is repaired. No broad qualification, calibration, packaging, or release proof has run.