Repository navigation
Conversation
…s step fail The weekly trend report has failed on both of its scheduled runs. On 2026-09-27 tri was not built yet (exit 127); on 2026-10-04 it was, and `tri stress --record` returned NotImplemented, as runStressTestCommand says every mode but --health does on purpose. On 2026-09-27 the step before it, "Run Full Stress Test", ended success over 15 compile errors in src/brain/stress_test.zig: `| tee` under the default bash -e reports tee's exit code. The brain-stress job had the same defect and was fixed; this copy sat in the tee-pipefail baseline. - Weekly job: both piped steps under `shell: bash`; record the score from `tri stress --health` instead of --record; drop --history; fix the artifact name, which `with:` took literally as "weekly-brain-report-$(date +%Y%m%d)"; let workflow_dispatch run it, so a fix can be checked before Sunday. - brain-health: drop "Record Health Snapshot", which wrote "Health snapshot recorded" to the summary on every run and recorded nothing. - brain-stress: drop the --record call; that job builds no tri binary. - tee_pipefail_baseline.txt: remove the weekly entry (77 known, 0 new). Closes #879 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
X Brain Health CheckScore: 100.0/100 X Brain is above merge threshold |
|
The weekly job's test, as promised in the description:
So the Done-when items of #879 that a run can show are met: the weekly job is green on Not shown by this run: that the stress step now goes red when the stress test fails. The run proves the step runs under 🤖 Generated with Claude Code |
🧪 Brain Stress Test ResultsScore: not measured (no stress test prints a Score: line) X Stress test passed |
1ffa2a1 to
97323e8
Compare
Summary by t27-beesThe PR fixes CI flakiness by replacing non-existent Worth a look
Read: 2 files, +33 -16, the whole diff. Head |
Closes #879. Part of #851.
What was red, and what was green that should not have been
The weekly trend report runs only on Sunday's schedule, and it failed on both of its runs. On 2026-09-27
triwas not built yet (exit 127). On 2026-10-04 it was built, andtri stress --recordreturnedNotImplemented. That is deliberate:runStressTestCommandimplements--healthand leaves every other mode an honestNotImplemented.On 2026-09-27 the step before it, "Run Full Stress Test", ended success over 15 compile errors in
src/brain/stress_test.zig. The cause:| teeunder the defaultbash -ereportstee's exit code. Thebrain-stressjob had the same defect and was fixed earlier; this copy was a known entry intools/tee_pipefail_baseline.txt.Changes
shell: bash(-eo pipefail)tri stress --record, failed every runtri stress --health | tee, undershell: bashtri stress --history(NotImplemented)weekly-brain-report-$(date +%Y%m%d)weekly-brain-report-${{ github.run_id }}if:workflow_dispatch--record || echo, then "Health snapshot recorded" in the summary on every run--record || echo "Record skipped"; this job builds no triMeasured locally
tools/check_tee_pipefail.pybefore pruning reported 0 new and 1 stale (the weekly entry). After--prune-baseline: 77 known, 0 new, rc 0.tools/test_check_tee_pipefail.py: 158 of 158.tools/check_run_syntax.py: 122 of 122 workflows read, 0 steps that do not parse.mainin the same 57 places, and adds none here.Not in this PR
workflow_dispatchon this branch is its test, and its result goes in a comment here. The pull-request run skips the weekly job.src/brain/health_history.zighasrecord,recentandtrend. Wiring--recordand--historyto it is the owner's call; see brain-ci weekly trend report: red on 2 of 2 scheduled runs, and its stress step went green over a compile error #879, step 2. On CI the jsonl would also have to survive between weekly runs.failed command: .../teston passing stress runs is not a failure. Zig 0.16's build runner prints a step's stderr with its command line whatever the result, and the benchmarks write to stderr. A comment above the step says so, so the next reader does not chase it.🤖 Generated with Claude Code