mirror of
https://github.com/vercel/workflow.git
synced 2026-09-14 19:59:43 +08:00
Alex Langenfeld
080c592567
## Summary & Motivation Job conclusions hide tests that pass on retry, and the evidence only lived in overwritten PR comments and 7-day artifacts, so recurring flakes couldn't be ranked against how often they ran. - retry sidecars are now named after the Vitest report they came from, so lanes, VMs, and worlds no longer overwrite each other when artifacts merge — and the aggregate comment can attribute a flake to an exact lane - `generate-e2e-flake-history.js` publishes a bounded 30-run history to gh-pages: one series per (lane, app, world, vm, platform) × test, carrying both an executed count and a passed-on-retry count so a rate has a denominator - the reporter always writes the sidecar, which is what lets a clean run be distinguished from a lane that reported no retry telemetry at all ## Test Plan Unit tests added for dimension parsing, denominators, schema validation, and the 30-run window; verified every current report filename maps to explicit dimensions, and a one-run history built from a full artifact set came out around 100 KiB.