safe-global / safe-client-gateway / 29007876698

90%

main: 89%

push

github

fix: make flaky-test analyzer Vitest- and rerun-aware, correct data from Jun 5

The log parser only knew Jest's failure format, so every failure after
the Vitest migration (#3136, Jun 23) parsed as "no failed tests" and
recent weeks falsely read 0% flaky. Additionally, flakes recovered via
"re-run failed jobs" were invisible at any point in time: the runs API
reports only the final attempt, so fail->rerun->success looked like a
plain success. Verified against all 230 post-cutover commits: not one
had separate pass+fail runs — same-run retries are how flakes surface.

- parse Vitest "FAIL <project> <path> > <test>" blocks and
  "(N tests | M failed)" summaries (ANSI/timestamp stripped first);
  keep the Jest matcher for pre-migration and lagging-branch logs
- for successful runs with run_attempt > 1, fetch jobs across all
  attempts and record the earlier failed attempt against the same SHA
- retry transient gh api failures
- data re-processed from the Jun 5 snapshot (dc890a1b8); recent weeks
  now read 5.6% / 3.7% / 2.0% instead of 0%, consistent with Datadog
  (unit 2.94%, integration 0.99% past week)
- file reformatted with the repo biome config (style-only churn)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

3077 of 3879 branches covered (79.32%)

Branch coverage included in aggregate %.

14759 of 16037 relevant lines covered (92.03%)

528.25 hits per line