All notes
Process

Two 'pre-existing failures' rot the whole suite

Two failing tests were skipped as 'pre-existing' for three weeks, surfacing as a recurring risk in every daily pulse. A permanently-quarantined failure trains the team to ignore red and drains the whole suite of signal — fix it or delete it with a documented reason.

Two failing tests in Whet’s curation backend — RecipeRegistryTest:13 and Scraper::BackfillTimelineJobTest:219 — were skipped as “pre-existing failures” on 2026-05-15 and were still skipped three weeks later, unchanged and untriaged. Every daily pulse in that window logged them as a recurring risk. The pulse named the failure mode exactly: sustained avoidance turning the suite untrustworthy. The label was never a status. It was a decision to stop believing the green checkmark.

”Pre-existing” is a verdict, not a description

Calling a red test “pre-existing” sounds like a neutral observation — it was already broken, not my doing. It isn’t neutral. It’s a standing decision to ship past a known failure indefinitely, renewed silently every run. The word launders a choice into a fact about the past. Two tests carried that label for three weeks with no triage attached, which means nobody ever decided anything; the default just persisted.

Quarantine doesn’t contain, it spreads

A skipped failure feels isolated: two tests out of a suite, walled off, the rest still green. It doesn’t stay walled off. Once the team learns that a red mark can mean “known, ignore it,” every red mark inherits the ambiguity. Was this one real, or is it another RecipeRegistryTest:13? The signal that a passing suite is supposed to give — this code is safe to build on — degrades across the whole suite, not just the two tests. That’s why the pulse escalated it as a project-level risk and not a two-line chore.

Fix it or delete it — with a reason

An indefinitely-quarantined test is worse than no test. No test costs nothing; a permanently-skipped one costs the credibility of every test around it while pretending to still be coverage. There are only two honest exits: fix the failure, or delete the test with a documented reason someone can later argue with. Both close the decision. skip leaves it open forever. The tell is the calendar — when the same skip appears in three weeks of consecutive pulses, the suite has already stopped telling you the truth.