[teamleaderleo]

Problem

When a PR's merge-base is older than a base fix, the PR still shows the red that the fix removed. The release-notary and workflow-guard tests are the usual case: test_every_part_of_bun_run_check_runs_in_the_warn_branch (tests/test_cmux_next_checks_workflow.py) and test_cmux_next_follows_the_committed_target_graph (tests/test_car_guards.py). base_red_excuse.py only excuses "the same failure on the base". The newest base passes those tests, so it calls the red the PR's own. next-autoland and the merge guard read it the same way, and reviewed PRs get stranded: #27584, #27918, #27905, #27775 and others.

Rule

When the newest base verdict for a red job is a pass, a failing test counts as fixed on base (not the PR's own) only if all of these hold:

• (a) The newest base run with a verdict for the job passed it, at a commit strictly after the PR's merge-base (compare merge-base...tested is ahead).
• (b) The job is a declared guard job (GUARD_JOBS): the cmux-next checks job, or a workflow-guard-tests / release-notary job. The failing test is a tests/*.py guard test, the failed step's own script runs that file, the step printed a test-run summary (Ran N tests or pytest's closing line), and the log names the test's file: either a pytest id or the test's own unittest traceback frame. Every failed step of the job must be such a test, or the checks job's own red check: rollup step naming no more reds than there are failed guard-test steps. The PR's diff from its merge-base (compare base...head, refused at 300 or more files) must touch neither the test's file nor anything it reads. Only a test whose file is declared in GUARD_READS qualifies, and an undeclared test is refused:
  • GUARD_READS_DEFAULT always applies: .github/, scripts/, tests/ and the release manifests (top-level files, release-notes/, homebrew-cmux/, cmux.xcodeproj/, Resources/, ghostty-next, vendor/).
  • GUARD_READS[file] adds the product paths that guard reads. test_cmux_next_checks_workflow.py declares Packages/, webviews/, cmux-tui/ and Native/ (the second review found it reads the Cargo manifests). test_car_guards.py declares Packages/, webviews/, cmux-tui/ and Native/.
• (c) The run is whole. A crash or a cut-short run never counts; that is still refused before this path is reached.

Any other test job (Swift, cargo, webview) is never fixed on base, because its tests read product code.

The audit line says so explicitly:

excused '<job>': fixed on base (passes at <sha> in feat-cmux-next run <id>, after the merge-base <sha>; PR does not touch tests/test_car_guards.py, .github/, scripts/, tests/, ...) (<run url>): <tests>

A refusal now gives the reason it is not fixed on base. It still starts with '<job>' is not red on <base> run <id> (success).

Risk: this is CI-trust code. The first review (gpt-6-astra high, SHIP at dcc2d01) reproduced a P1 against the original fallback ("touches nothing under .github, scripts or tests"). tests/test_docs_deploy_auth_guard.py reads web/vercel.docs-channel.json, so a PR that broke that config was excused. Undeclared tests are now refused instead. The third review (BLOCK at d35d239) found that any job printing a FAILED tests/x.py:: line could be excused. Only declared guard jobs qualify now, and only for steps that run the file and finish a test run. To let another guard be excused, declare it in GUARD_READS once its reads are checked.

Tests

tests/test_base_red_excuse.py gets a new FixedOnBase class covering:

• the excused case
• the PR touching the test's file, a default path, or a declared path
• an undeclared guard, which is refused
• a base pass that is not after the merge-base
• a crash and a cut-short run
• a failed step with no guard test
• a Swift job, which is never excused and never calls compare, including one that prints a guard-test line
• a step whose script does not run the test file
• a pytest line with no test run behind it
• a diff that is too long
• the checks-job rollup step, including a rollup with more reds than guard tests
• pytest ids

The existing Excusing tests still pass.

🤖 Generated with Claude Code

Agent session: agent session

[teamleaderleo]

Review (codex gpt-6-astra high, router-review): SHIP at 00085d3

Follow-ups: none

Earlier heads: SHIP at dcc2d01 (P1 follow-up, fixed in 627b0b3), SHIP at 627b0b3 (P2 follow-up, fixed in d35d239), BLOCK at d35d239 (non-guard jobs could be excused, fixed in 00085d3).

[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 00085d3: 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=pr30467 -f ref=ci/base-red-fixed-on-base -f platform=macos -f backend_mode=local

Shipped in:

• cmux NEXT nightly 1.0.0-nightly.4130052256501 (28aa638015)

[blacksmith-sh]

Found 9 test failures on Blacksmith runners:

Failures

Test	View Logs
/tmp/tmpo9rawp0d/summary-missing.json (live):/summary-missing.json (live):	View Logs
/tmp/tmppd5h8502/summary-fail.json (live):/summary-fail.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
test_no_suite_runs_page_javascript_on_a_built_bundle (main.BundleInputs.test_no_sui te_runs_page_javascript_on_a_built_bundle)/ test_no_suite_runs_page_javascript_on_a_built_bundle (main.BundleInputs.test_no_sui te_runs_page_javascript_on_a_built_bundle)	View Logs
the committed export failed: check-daemon-capabilities-fresh: plans/cmux-next/ daemon-capabilities.json is stale:	View Logs
the committed export failed: check-daemon-capabilities-fresh: plans/cmux-next/ daemon-capabilities.json is stale:	View Logs

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