[teamleaderleo]

Follow-up to #26798. scripts/ci/base_red_excuse.py (used by gh-merge-green) and next-batch decide whether a red on a PR is "the same red as the base". The fleet Release compile's red while the controller was wedged is cmux-next/compile is not a pass for <sha> (no result). Error lines were compared with only digits masked, so the PR's line and the base's line never matched because each names its own sha. Masking shas alone would have been wrong, though: it would excuse a compile that compiled nothing, and a Swift PR could land unbuilt.

The rule this PR enforces: a fleet compile "no result" is an infra non-verdict, not a test failure. It is never excused as a base red, even when the PR and base show the identical normalised error. It may be set aside only when another job at that same head gave a real compile pass: the Mac Release compile fallback green, or cmux-next/compile success.

Change

• base_red_excuse.judge: a failed job whose log says it gave no compile verdict is refused, whatever the base shows. It fails closed: of compile-status's own error lines, only cmux-next/compile is not a pass for <sha> (fail) is a real compile red. Every other state (no result, unavailable, mismatch, unreadable, pending, last: ...), could not post the pending status and bad head sha count as no verdict, as does the app compile's No build-fleet verdict for ... from ci(cmux-next): a build-fleet app compile with no verdict fails, never green #26798. Only the job's own ##[error] lines count, never the script text the log echoes. It is set aside, with an audit line ignored '<job>': ... compiled <sha> and passed, only when one of these holds at the same head:
  • the newest attempt of the Mac Release compile job (cmux-next Release compile (Xcode 26)) at that exact sha concluded success, with its Release compile or Reuse that pass step green. The newest attempt decides, so an older pass doesn't count over a newer failure. release-compile-fast-admission, a Linux wrapper with the same check name, has neither step and never counts. A check run whose sha is null or different doesn't count either;
  • the newest cmux-next/compile commit status is success.
• next_batch.expand_job: such a job, or an unreadable log of a compile job that can end without a verdict, is keyed by its own job id (no compile verdict (job N, never a base red)), the same way crashes already are, so a base probe never excuses it.
• Compared error lines now mask commit shas before digits: a full 40-hex sha, or 7–39 hex characters right after a word that names a commit (commit, sha, tree, revision), any hex digits at all there. Any other hex id, such as an artifact hash, is compared as is. A real red that differs only by the commit it names is still excused. Nothing else gets looser: steps, test names, crash and truncation rules are unchanged.

Tests

• tests/test_base_red_excuse.py NoVerdictCompileIsNeverABaseRed:
  • no-verdict with nothing else compiling the head is refused, even when the base has the same "no result" line;
  • the same holds for the app compile's "No build-fleet verdict" line;
  • it is ignored when the Mac Release compile passed at the head, but a skipped run or one from another app does not count;
  • it is ignored when cmux-next/compile succeeded, but not when the newest status is failure;
  • a real red differing only by its embedded sha is excused;
  • a different error is not;
  • the admission job of the same name, an older pass under a newer failed attempt, and a skipped run, a run from another app or one with a null or different sha are not passes; a reused pass for the same inputs is;
  • a real fleet red whose log echoes the no-verdict script text is not set aside;
  • other hex ids (artifact deadbeef1 vs artifact abcdef12, for deadbeef1 vs for abcdef12) are not equated;
  • every compile-status state except (fail) is no verdict, and a real (fail) is compared with the base;
  • next-batch: an unreadable no-verdict compile log is keyed by its own job.
• tests/test_next_batch.py: a no-verdict compile is keyed by its own job.
• On today's code, most of the new excuse tests fail; all pass on this head. Locally, test_base_red_excuse, test_next_batch, test_next_batch_groups and test_test_reds all pass (244 tests). Checked against the real logs too: both "no result" fleet Release compile jobs and the "No build-fleet verdict" app compile job are recognised, and the [launch-polish] Add native human setup guide #26741 green log is not.

Risk: low. Every change except the sha mask makes the excuse stricter. The sha mask only ever applies after a no-verdict compile has already been handled.

Agent session: agent session

🤖 Generated with Claude Code

[teamleaderleo]

Owned by an agent session · session

[github-actions]

cmux status _{(one comment, edited in place; each line is updated by the job that owns it)}

Review 78e86fa: HQ feed card · gallery before/after runs · checks

App of this head (app.zip, kept 1 day)

gh workflow run reload-build.yml -R manaflow-ai/cmux --ref feat-cmux-next -f tag=pr26816 -f ref=base-red-sha-normalise -f platform=macos -f backend_mode=local

red: cmux-next checks (god files, concurrency, crash safety, l10n); NEW: red check: Package dependency lists sorted is not red on the base 382bbd6; do not merge until green; see lane-rules

[teamleaderleo]

Bead cx-reoc9l created for this PR by the cmux coordinator.

[cursor]

Bugbot needs on-demand usage enabled

Bugbot uses usage-based billing for this team and requires on-demand usage to be enabled.

A team admin can enable on-demand usage in the Cursor dashboard.

[blacksmith-sh]

Found 11 test failures on Blacksmith runners:

Failures

Test	View Logs
/tmp/tmp6hx0x41z/summary-fail.json (live):/summary-fail.json (live):	View Logs
/tmp/tmposjk8per/summary-missing.json (live):/summary-missing.json (live):	View Logs
[new_workspaces] debug.new_workspace timed 0 of 50 new workspace rows/ [new_workspaces] debug.new_workspace timed 0 of 50 new workspace rows	View Logs
[new_workspaces] the window draws no frames/[new_workspaces] the window draws no frames	View Logs
[tabs] debug.tab_switch timed 0 of 40 switches/ [tabs] debug.tab_switch timed 0 of 40 switches	View Logs
[tabs] the window draws no frames/[tabs] the window draws no frames	View Logs
build_only must be a conjunctive exclusion on the helper, signing, and publication jobs , and must not gate the unsigned app build/ build_only must be a conjunctive exclusion on the helper, signing, and publication jobs , and must not gate the unsigned app build	View Logs
nightly.yml must derive the embedded key and sign every appcast through the helper/ nightly.yml must derive the embedded key and sign every appcast through the helper	View Logs
test_prepare_runs_on_linux_and_executes_no_mach_o (main.NightlyPooledRoutingTests.t est_prepare_runs_on_linux_and_executes_no_mach_o)/ test_prepare_runs_on_linux_and_executes_no_mach_o (main.NightlyPooledRoutingTests.t est_prepare_runs_on_linux_and_executes_no_mach_o)	View Logs
test_the_next_feed_keeps_its_newest_nightly_item_before_it_goes_live (main.Workflow Tests.test_the_next_feed_keeps_its_newest_nightly_item_before_it_goes_live)/ test_the_next_feed_keeps_its_newest_nightly_item_before_it_goes_live (main.Workflow Tests.test_the_next_feed_keeps_its_newest_nightly_item_before_it_goes_live)	View Logs

...and 1 more test failures. View all on Blacksmith

^{Need help on this PR? Tag @codesmith-bot with what you need.}

[github-actions]

The feat-cmux-next push run https://github.com/manaflow-ai/cmux/actions/runs/40693729434 at 3df9106 failed: cmux-next checks (god files, concurrency, crash safety, l10n).
Those jobs last passed at b02480c. This pull request is one of 2 merged since: #26816, #26700.
If the failure is in your change, fix forward on feat-cmux-next. Pull requests run only the tiers their paths reach (docs/ci/cmux-next-tiers.md); label a batch PR full-ci to run them all before merging.

[github-actions]

Shipped in:

• cmux NEXT nightly 1.0.0-nightly.4069874506401 (72bb01dd5b)
• cmux NEXT nightly 1.0.0-nightly.4070477034801 (04673a0583)