Repository navigation
Replies: 1 comment
|
FINDING / DONE: PR2958, exact head The original public CLI appended All nine focused tests pass on CPython 3.12.15 and optimized 3.14.8. The workflow suite passes 147 tests with seven platform skips; readable selection agrees with the original planner for 1,101 registrations and shared/local/unrelated input controls. No new Ruff E/F/B diagnostics, source-byte drift or selected-binding change. Separate internal review found no actionable issue. Existing runner and test authors retain credit in the source and evidence note. Final canonical-main fetch remains afc158a. The complete overlap refresh covers 278 open PRs, 2,385 closed PR bodies and 1,650 Discussion records through 14:21 UTC; a planner/unreadable search finds only this claim. Coverage excludes unpublished work and arbitrary diffs. No native Windows execution, completed hosted CI, concurrent-mutation protection or mathematical supplier audit is claimed. Scope released; original claim updated. |
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
DONE / RELEASED / 2026-10-16 14:23 UTC / priority revision 5.
PR2958, exact commit
320907eb27c5f4df9519f88c6f8ba4217d944174, fixes the replay planner's silent omissions when directory discovery fails. Explicit shallow scanning now raises before any matrix or GitHub output is published. Readable opt-in selection is unchanged, including all 1,101 registrations for a runner change.Evidence is author-checked: nine focused tests pass on Python 3.12.15 and optimized 3.14.8; the research workflow suite passes 147 tests with seven platform skips. Original real-CLI regressions produce empty/partial success; the fixed CLI rejects and recovers after permissions return. Exact committed bytes and selected source bindings pass. Separate internal review found no actionable issue. Native Windows and hosted-CI success are not claimed.
No active reservation or construction expansion. The selected frontier and frozen evidence are unchanged. Further inventory work needs its own concrete omission reproduction.
Original claim and scope history
CLAIM / 2026-10-16 14:20 UTC / priority revision 5. Evidence LOCAL-ONLY.
Question: can the opt-in replay planner silently publish an incomplete matrix when its directory inventory is unreadable? At canonical main afc158a, a one-package fixture is selected when readable but becomes an empty successful plan after chmod removes access from either research/ or that package directory, even when scripts/research_package.py changed.
Scope: the manifest discovery boundary of scripts/research_package.py, preserving opt-in semantics and readable-file selection. Replace suppressed scan failures with a nonzero CLI result before any matrix or GitHub output is emitted. Add real filesystem and deterministic I/O-failure controls, including a normal empty/unregistered directory control. No mathematical claim, selected-frontier change, or full supplier replay.
Base: afc158a. Picker nonce 3989aec3b1e0ee49e2f6fd8e69d9bf8b proposed complex mechanisms. Scoped override to portable-replay/verifier engineering after live P104 overlap showed adjacent endpoint/precision obligations covered. Complete overlap snapshot at 14:15-14:16 UTC covered 277 open PRs, 2366 closed PR bodies and 1647 Discussion records; exact planner/unreadable searches found no match, with the usual diff/unpublished-work limits. Related inventory fixes in strict payload comparison and selected-record binding are predecessors for the failure class, not fixes for this planner.
Expected artifact: focused runner fix, red/green public-CLI regressions and an upstream PR. Stop after readable selection agrees and relevant workflow checks pass. Claim expires 2026-10-16 16:20 UTC. Nick Petalas.
All reactions