Repository navigation
Require history-backed navigation checks - #779
Conversation
|
Additional scoped execution at local commit |
|
The exact-head navigation run |
Preserve Main #779 and every decision row. The current Main decision ledger remains an exact prefix; the two recovery decisions follow it unchanged. The inventory tool, its tests, recovery guide and plan are byte-identical to the reviewed candidate. No proof or scientific disposition changes. Actual composition: OpenAI/Codex. Local equivalent merge 3579831c has the identical tree.
The navigation workflow used a shallow checkout, so historical-source reader tests reported skipped checks while the job succeeded. Fetch full history and run the two history-backed reader classes as an additional fail-closed step: the child must succeed, each selected control must report
ok, and the final unittest summary must have a positive count and no skips. The existing discovery run and test bodies remain intact.Local verification on the committed candidate: 25 focused workflow/discovery/guard controls passed in each Python mode; a real full-history checkout passed 20 selected tests, while a real depth-one checkout recorded 12 skips and was rejected. Nine designated workflow mutations were killed in both modes, with traceback code-origin controls rejecting subject, helper, and nested-callback assertions as oracle failures. The initial guard's false acceptances of forged earlier summary lines were reproduced and corrected. Evidence: Library file
libfile_fde3_f-f232855b2dc081919315ab3c31a32f56_59a0f9f234e48191bc19fddcc0f80648, SHA-2566fa1b43e564baa27270e311f44b56cd2e580b2564c8253805789ab17f4c39fcf.This is a scoped workflow repair, not a full regression or CBZ-0 qualification. Other observed skip categories remain unresolved, and hosted checks and artifacts must be inspected at the current head. Canonical status records the evidence and the existing checkpoint/archive version split.
Hosted qualification at tested merge
ca1154a06b4ad3afc31227846c87479b1361ff25: navigation artifact13695969227records 2,469 discovery tests with 25 skips and a separate 20-test history-reader run without skips. All five PR-triggered workflows returned success; the protected merge was rejected because requiredpublic-intakehas not yet reported. Its normal trusted-base check must run and succeed before integration. The five sandbox, two browser and two Lean cases skipped by navigation also ran successfully in their separate hosted jobs; these observations do not turn the skipped discovery into full regression 100%.A separate adverse finding concerns the existing node-status freshness mutation grader: a subject-origin
AssertionErrorwas wrongly counted as a kill in both child modes. Evidence is retained inlibfile_fde3_f-606b8b159414819198b7a213ca93d32c_59a0f9f234e48191bc19fddcc0f80648, SHA-256af7e53894196fd12ad6b0e42e0c022656358f4aa4e5f556c21b9b3b0c92f5c74; a disjoint repair is underway. This PR does not change that grader or claim its mutation evidence is provenance-qualified. The prior trustedpublic-intakeruns failed to start during GitHub's scheduledpull_request_targetbrownout ending 18:00 UTC; all failed runs and the rejected merge remain recorded.