Skip to content

reviews.md

Latest commit

 

History

History
21 lines (13 loc) · 2.54 KB

reviews.md

File metadata and controls

21 lines (13 loc) · 2.54 KB
Raw

PR claims and documentation

Part of the coding standards. Check the PR body and commit messages against the diff, the same way you check the code.

Claims match the diff

PR bodies and commit messages are evidence. The second-pass reviews on #34 to #38 blocked two PRs for claims the diff contradicted.

  • A claim of scope must survive the diff. "Nothing here is staging-dependent" is false when the diff removes the staging flag (#37). "On every page" is false when the diff edits one file (#38).
  • A deletion shows the search behind it, covering dynamic imports, string-built paths, e2e/, and page.evaluate access that typecheck cannot see (#34, #36).
  • "No behavior change" holds for every input. Response.json(undefined) throws where JSON.stringify did not (#36).
  • Test counts in the commit and the PR body agree and come from the final head (#35).
  • Proof that used stubs says so: "This proves client dispatch only; it is not authenticated provider proof." (#65)

Documents change with the code

The owning document changes in the same PR as the behavior it describes (docs/README.md names the owners). Reviewers found a plan with two statuses for one feature, a release checklist missing test:edge, a migration marked applied in one file and pending in another, and new ports with no documentation (#17, #26, #63). Commands in plans run on the pinned CLI version (#29).

The diff holds only this change

A diff full of unrelated files usually means the branch was cut from a stale local main. A branch like that once put 20 plan files into a PR, and all three review findings landed on them (#29).