ci(brain): skip the health report when the run itself was cancelled

[gHashTag]

Closes #881. Part of #851.

What was wrong

brain-health-report in brain-ci.yml ran on if: always(). It reported DAMAGE and ran exit 1 unless all six needs were success, and on a pull request it then posted "🚨 CRITICAL: Brain Health Below Threshold … This PR cannot be merged".

A cancelled need is not success, so when a newer push cancelled a run, the report treated the cancel as brain damage. On #870:

• run 37228527559 was cancelled while its stress test was running (20:58:18Z);
• the report ran at 21:10:15Z, failed, and posted this comment;
• the replacing run 37234155119 was green in every job.

main's brain-ci.yml has no concurrency: block, so today this happens only on a manual cancel. #870 adds cancel-in-progress for pull requests, which would make it happen on every push during a run.

The change

One line: if: always() → if: ${{ !cancelled() }}, with a comment. cancelled() is true only when the workflow run was cancelled, so the report still runs after a need fails, and still goes red and posts.

Scope

Of the 10 jobs on main with a job-level always(), this is the only one that requires success from its needs. fpga-ci.yml and fpga-regression.yml have summary jobs that exit 1 only on == 'failure', which a cancelled need does not match. No step-level always() step posts a comment. Scan: a text scan of .github/workflows at 57de4a1.

Checked locally

• tools/check_run_syntax.py: 122 of 122 workflows, 0 steps that do not parse.
• tools/check_tee_pipefail.py: 78 known, 0 new.
• research/audit_workflow_paths.py: 0 dead paths.
• git merge-tree is clean against ci: give the 102 push workflows a dispatch-safe concurrency group #870, ci(brain): stop calling tri stress --record, and let the weekly stress step fail #880, ci: run nine workflows once per PR commit, on every path their push run watched #876, ci: run the nine document gates on push to main only #867 and fix(gates): green doc-refs and orphan-artefacts on main after #510 #872.

Shown by CI, 2026-10-04

Four brain-ci runs. The two cancelled ones were cancelled at the same point, while the stress test was running, which is where #870's run was cancelled.

run	ref	what happened	📋 Brain Health Report
37239407266	main (always())	cancelled during the stress test	ran and failed: [ "cancelled" == "success" ] → DAMAGE → exit 1. This is the defect, reproduced.
37239404431	this branch	cancelled during the stress test	never started: 0 steps, no runner. GitHub lists it as cancelled (not skipped, as I first predicted).
37239392265	this PR	not cancelled, all needs green	ran, success
37239915168	exp/report-fail-ab = this branch + a forced exit 1 in Build Check	not cancelled, one need failed	ran and failed: [ "failure" == "success" ] → DAMAGE → exit 1. A real failure is still red.

The two dispatched runs could not post a comment, because the notification step runs only on pull_request. What these runs show is the report's result, which is the step's precondition.

exp/report-fail-ab is a throwaway branch with one commit (88bc1eb). It is not for merge.

Not measured

• The case where a need hits its own timeout-minutes. The workflow is not cancelled then, so the report runs, but I did not check which result the need gets.
• A cancel that lands before detect-build finishes. The report on main then takes its "skipped" branch and passes (inferred from run 37227380293, not reproduced). This PR does not change that branch.

🤖 Generated with Claude Code

[github-actions]

X Brain Health Check

Score: 100.0/100
Status: 🟢 HEALTHY
Threshold: 80/100

X Brain is above merge threshold

[github-actions]

🧪 Brain Stress Test Results

Score: not measured (no stress test prints a Score: line)
Status: 🟢 PASS
Threshold: 270/300

X Stress test passed

[tri-claw]

Summary by t27-bees

This change prevents the Brain Health Report job from running when the workflow run itself is cancelled (e.g., superseded by a newer push). Previously if: always() caused the report to execute after a cancellation, incorrectly treating cancelled dependencies as failures and posting a blocking "CRITICAL" comment on the PR.

Worth a look

• .github/workflows/brain-ci.yml:499 — The new condition if: ${{ !cancelled() }} skips the report when the run is cancelled, but it still runs when a need fails (since cancelled() only reflects the workflow run state, not individual needs). This matches the intent, but verify that a need hitting its own timeout-minutes (which does not cancel the workflow) still triggers the report as expected.

---

_{Read: 1 files, +4 -1, the whole diff. Head facf048. Model: openai-compatible/nvidia/nemotron-3-ultra-550b-a55b. Commands: @t27-bees review | summary | help | pause | resume. Free for every repository.}