Repository navigation
Sweep fix: Lab scores qa-bench runs that exit 1 or 3 from write_report - #2106
Merged
Merged
Conversation
transcripted-qa-bench.sh's write_report writes qa-report.md, then exits 1 when a step FAILed or 3 when a step was skipped or held (always the case with --skip-build). LabExperimentRunner treated any nonzero exit as a failed process and never ran QAResultsAnalyzer, so those runs lost their whole breakdown. Now exit 1/3 gets analyzed only when it came from write_report: qa-report.md must exist (results.tsv exists from the script's first line and can't prove that), and exit 1 also needs a FAIL row. FAIL rows still become hard gates; exit 3 adds a warning. Timeouts, other exit codes, set -u aborts, and the early cd failure still fail as before.
r3dbars
marked this pull request as ready for review
October 8, 2026 10:22
r3dbars
added
the
waiting-on-human
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Split out of #2101 (one PR per bug). Confirmed bug from the Haiku sweep.
What was wrong
transcripted-qa-bench.sh'swrite_reportwrites qa-report.md and results.tsv, then returns 1 on any FAIL row and 3 on any SKIP or hard-WARN row.--skip-buildalways records a SKIP, so it always exits 3.LabExperimentRunnertreated any nonzero exit as a dead process and returned beforeQAResultsAnalyzerran, so the run showed as Failed with no metrics or step breakdown.Failing tests
QABenchExitCodeTestsruns the real runner against a fake repo whose qa-bench script writes the files and exits with a given code.testQABenchHoldExitStillScoresSkippedRun,testQABenchHoldExitIsNotReportedAsAPassandtestQABenchFailExitKeepsStepBreakdownwent red ("Experiment process exited with code 3."), then green.testQABenchAbortAfterPassingStepsStillFails(exit 1, PASS rows only, no report; was scored passed),testQABenchAbortWithEmptyResultsStillFails(was warning) andtestQABenchFailExitWithoutFailRowsStillFails(was passed) went red, then green.testQABenchCrashWithoutReportStillFailsandtestQABenchUnknownExitStillFailsEvenWithReportguard the failure side.Fix
For the QA and transcription-corpus benches, an exit of 1 or 3 gets analyzed only when it came from
write_report:cd "$REPO_ROOT" || exit 1,set -uaborts).write_reportstill fail like before.Checks
swift test --package-path Tools/TranscriptedLab --jobs 4: 16 tests passswift buildfortranscripted-labandTranscriptedLab: passTools/TranscriptedLab/script/build_and_run.sh --verify: pass, run by handbash check.sh: everything else passed. It reportsbuild_and_run.sh --verifyas BLOCKED (invalid-command) becauseagent-check.pywon't run a bare-path matrix command. That was already true on main, so this check isn't covered bycheck.sh.Independent review: approve with nits (3 rounds, reviewed as part of #2101). No open nits for this fix.
🤖 Generated with Claude Code