The release gate judges the latest finished run
A passing re-run now clears an earlier failure, a run still in progress never counts as a pass, and tests only count for the version of the code they ran on.
The release gate decides whether a change can ship. Three fixes make its answer match what actually happened:
- The latest finished run decides. The gate blocked on any failed run in its 14-day window, so push, fail, fix and re-run left a pull request blocked for two weeks. Now each workflow is judged by its latest finished run. A passing re-run clears the earlier failure. A re-run still in progress, or one that was canceled, does not.
- No pass without evidence. A CI check seconds after a push could be told "release ready" while the only run was still running. Now at least one run must have completed, and a run whose AI review never ran is refused.
- Tests count for the code they tested. If the change is edited after its tests ran, those results no longer block or pass it, and the gate says this version is untested. (This one landed on 8 October.)
See quality gates for what the gate checks.