Designing flake triage that engineers actually open

Flaky UI tests die in silence when artifacts read like random hashes. We start by aligning artifact names with user journeys—checkout happy path, account recovery edge—and tie each to the owning squad in the README footer.

The second paragraph is about ritual, not tooling. A fifteen-minute weekly flake stand-down with rotating note-taker beats a monthly spreadsheet nobody trusts. Keep the agenda boring: top three flakes, suspected environmental cause, next experiment, explicit owner.

Finally, publish a one-page decision tree for reruns versus quarantine. Engineers respect boundaries when the document admits uncertainty instead of pretending every rerun is free. Close the loop by posting incident-style timelines for flakes that turned into real defects; that single habit keeps triage meetings attended.