[codex] Validate changelog matrix before sweep reuse

[Oseltamivir]

Summary

• remove the strict immutable-history check-changelog behavior from run-sweep.yml
• replace it with a setup-compatible check-changelog job before sweep reuse
• require exactly two validations: a final newline and successful process_changelog.py matrix generation
• remove the obsolete strict validator CLI mode and historical strict-validation CI step from Add failed-ingest recovery and pre-reuse changelog validation [skip-sweep] #1821
• retain the existing reuse, setup, merge-recovery, and sweep-label conflict behavior

Why

PR #1852 removed the duplicate associated-PR validation from the test workflow, but the production workflow still enforced immutable historical entries. That caused PR #1833 to fail with entry 497 changed.

A newline-only precheck is insufficient. Run 27729613948 failed after PR #1717 merged because conflict resolution changed the existing PR #1798 line; process_changelog.py then rejected the diff before generating a matrix. The replacement check executes that same processor before the reuse gate, so malformed conflict resolutions cannot bypass setup via sweep reuse.

There are no PR-link or other exceptions: a PR must have a final newline and generate a valid matrix to pass this gate.

#1821 cleanup

No complete file from #1821 is redundant without removing active functionality. The recovery command/helper, merge-preparation helper, reusable-artifact validation, and focused test workflow all still have live callers. This PR removes the dead strict-validation branch and its historical CI step instead.

Validation

• actionlint .github/workflows/run-sweep.yml .github/workflows/test-changelog-gate.yml
• bash -n utils/merge_with_reuse.sh
• Python compile check for validator, merge preparation, and recovery helpers
• focused test suite: 94 passed
• regression replay of run 27729613948: rejected with the expected deleted PR [NV]Add GLM-5 NVFP4 GB300 disagg-non-mtp TRT-LLM benchmarks via Dynamo  #1798 line
• replay of PR [codex] Update MiniMax M3 B200 FlashInfer image #1833: accepted and matrix generated

---

Note

Medium Risk
Changes production sweep and reuse gating for perf-changelog.yaml; push-to-main no longer runs the old strict validator job, relying on setup's processor instead.

Overview
Replaces the strict immutable-history check-changelog gate in run-sweep.yml with a PR-only check that matches sweep setup: trailing newline plus successful process_changelog.py matrix generation via validate_matrix_compatible_change. Drops has-additions / metadata-only outputs and no longer blocks setup on “no new entries”; push-to-main skips check-changelog while setup still runs process_changelog.py.

reuse-sweep-gate no longer requires appended entries—only a successful matrix check on synchronize before evaluating /reuse-sweep-run. merge_with_reuse.sh waits for check-changelog then all PR checks before squash-merge. Docs (README, KLAUD_DEBUG) and gating tests are updated; the historical strict-validator CI step is removed from test-changelog-gate.yml.

^{Reviewed by Cursor Bugbot for commit 11f5456. Bugbot is set up for automated code reviews on this repo. Configure here.}

[claude]

Claude finished @Oseltamivir's task in 2m 40s —— View job

---

PR Review: Validate changelog matrix before sweep reuse

• Read changed files and understand the changes
• Check for bugs, logic errors, and blocking issues
• Post review findings

LGTM — no blocking issues found.

The refactor correctly replaces the immutable-history check-changelog gate with a lighter --matrix-compatible precheck (final newline + process_changelog.py matrix generation). The key behavioral changes are well-reasoned:

1. check-changelog now PR-only — push-to-main events skip it, and setup accepts skipped via the new success || skipped clause. Pushes still get validated by process_changelog.py inside setup itself.

2. has-additions output removed — the reuse gate and setup no longer condition on it, which removes the metadata-only skip path for push events. This is an intentional simplification; setup's own matrix generation handles empty-matrix cases.

3. Conflicting sweep labels moved to setup — functionally equivalent since a check-changelog failure already blocked setup. Moving it to setup keeps check-changelog focused on changelog validation only.

4. merge_with_reuse.sh uses gh pr checks --watch — replaces the custom wait_for_check polling loop with a single gh pr checks call that waits for all checks, not just check-changelog. Simpler and catches any check failure before merge.

The test suite is thorough — the exhaustive cross-product covers 1154 scenarios (down from 2312 after removing the has-additions axis) and the reference spec independently encodes the intended gating logic. The test_matrix_compatible_check_rejects_pr_1717_conflict_resolution test pinning real commit SHAs is a nice regression guard.

Re: CodeQL dual-import finding — the import validate_perf_changelog as validator on line 8 is needed for monkeypatch.setattr(validator, ...) in the new tests. The existing from validate_perf_changelog import (...) serves the pre-existing tests. Both imports are load-bearing; the CodeQL warning is a false positive in this context.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 60928c8. Configure here.}