From c1ae691e9703b22d0f40cfd3e16700bc584e5d19 Mon Sep 17 00:00:00 2001 From: Mike Grier Date: Mon, 14 Sep 2026 16:12:52 -0400 Subject: [PATCH] docs: sweep the change set, not only the file you noticed the defect in Earned by the review of PR #90, where two defects survived a sweep I had already done. Peeling a probe onto `main` brought prose written against an older state. I swept the design notes for links whose targets no longer existed, found three, and fixed them -- then committed without asking whether the SOURCE FILES moving in the same commit had the same defect. They did: the module doc named the same absent checklist, and three sites named a queue module that had been renamed. A reviewer found both. CONTRACT INTEGRITY rule 3 is the natural home and already says "the reported site is a sample, not the population", but neither half of it fires here. Its trigger is a change to a stated CONTRACT rule, and a dangling link is not one; its population is the component and its dependents, reached by grep, not the commit in front of you. So this is a gap rather than a restatement, which is why it lands as one sentence inside rule 3 rather than as a second sweeping rule somewhere else -- two rules about sweeping would be exactly the restatement drift this section exists to prevent. Deliberately narrow, and worth saying so: this is mechanical and cheap, and it would have caught both prose defects in that PR. It would NOT have caught the same review's most valuable finding -- that enabling two non-default features on a workspace member removed the default-features build from CI -- which needed a reviewer reasoning about feature unification, and which no amount of grepping one's own diff would surface. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> --- .github/copilot-instructions.md | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/.github/copilot-instructions.md b/.github/copilot-instructions.md index 6bc6c13fd..d1e8f0905 100644 --- a/.github/copilot-instructions.md +++ b/.github/copilot-instructions.md @@ -1254,6 +1254,10 @@ files, 4 updated, 9 incidental mentions"). Do **not** fix only the site a review the reported site is a sample, not the population. This is cheap and it works: run once voluntarily on this repository, a sweep immediately found a stale site no reviewer had reported. +**And sweep the change set, not only the file you noticed it in:** when you fix a defect of any +kind in one file of a commit, grep the commit's other files for the same defect before committing +-- the file you happened to be reading is a sample of your own change too. + Two corollaries that have each already cost a review round: - **An analysis document never restates normative content.** An audit, review response, or design