pjandro/OmniRoute
1
0
Fork 0
mirror of https://github.com/diegosouzapw/OmniRoute.git synced 2026-07-26 09:52:11 +03:00
Code Issues Packages Projects Releases Wiki Activity

docs(quality): codify retry policy per runner + release-level drift rule (WS5.4/WS5.5) (#7107)

Browse Source
This commit is contained in:
Diego Rodrigues de Sa e Souza
2026-07-14 03:55:32 -03:00
committed by GitHub
parent 9767b7eb34
commit 131a48344c
2 changed files with 26 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
changelog.d/maintenance/quality-policies-ws5.md Normal file
View File

@@ -0,0 +1 @@
- **Docs**: `QUALITY_GATES.md` now codifies the per-runner test retry policy (Playwright 1 CI retry with trace; Vitest per-test explicit quarantine only; node:test never) with target flake SLOs, and the release-level ratchet-drift rule (combination drift on the pure tip is the release captain's to fix once on the branch — never pushed onto contributor PRs, never rebaselined per-PR)

25
docs/architecture/QUALITY_GATES.md
View File

@@ -173,6 +173,31 @@ pending implementation).
---
## Test Retry Policy (WS5.4, v3.8.49)
Retry is per-runner, never a global blanket — a blanket retry converts real regressions
into invisible flakes:
| Runner | Policy | Why |
| --- | --- | --- |
| Playwright (e2e) | `retries: 1` in CI only, with `trace: on-first-retry` | Browser/network timing is genuinely nondeterministic; one retry with a trace turns a flake into a diagnosable artifact |
| Vitest | NO global retry. A proven-flaky test gets an explicit per-test retry (visible in the diff, reviewed in PR) | Keeps the quarantine list in the repo, never opaque |
| node:test (unit) | NO retry, ever | A flaky unit test is a bug in the test — fix it, don't re-roll it |
Target SLOs once flake telemetry lands (WS5.2/5.3): <1% flake rate per test
("fix now" threshold), ≥95% pass rate per pipeline. Industry reference values —
recalibrate against our own measurements.
## Release-Level Ratchet Drift (WS5.5, v3.8.49)
When a ratchet (file-size, complexity, eslint warnings) regresses on the PURE release
tip — i.e. the COMBINATION of merges regressed it, and no single PR reproduces the
regression on its own branch — the fix belongs to the **release captain, once, on the
release branch**: prefer extraction/refactor; rebaseline only with the documented
justification entry. Never push combination drift onto a contributor PR, and never
rebaseline per-PR (that hides real regressions). Discriminate first: reproduce the
red against the pure tip in a probe worktree before assuming your PR caused it.
## Allowlist Policy
Every gate that cannot fail on pre-existing violations uses a frozen allowlist