finding(scripts): the same drifted counts objectui#7965 fixed on the doc page also sit in a gate docblock and a workflow header, where no pin reads them

[baozhoutao]

Found while implementing objectui#7965 (three drifted file/document counts on content/docs/guide/ci-cd-pipeline.md). Not fixed there: that card's file surface is the one doc page, and both sites below are under scripts/ and .github/workflows/, which the dispatch explicitly excluded. Filed rather than folded in.

Two more copies of the same numerals, outside content/docs

Both are prose comments. Both carry counts that were true when written and are wrong today, and neither has a declared measurement point — the same shape objectui#7965 was filed for.

1. scripts/check-shell-escape-residue.mjs:232 — "203 of today's 207 files"

 * ⛔ `minFiles: 3` for that row -- PER ROOT, never a whole-surface floor, which
 * is the shape that failed: on a day this four-file root reads ZERO the other
 * four roots still return 203 of today's 207 files, so any total floor is green
 * through the whole outage.

This is the gate's own docblock, and it is where the doc page's sentence was copied from (or the other way round — they are near-identical). The gate's own verdict line today, run on main 59a3a233d:

✅  check-shell-escape-residue: OK (5/5 root(s) resolved -- AGENTS.md: 1 file(s), 15 fence(s); CLAUDE.md: 1 file(s), 0 fence(s); skills: 16 file(s), 210 fence(s); .claude/skills: 4 file(s), 20 fence(s); content/docs: 184 file(s), 1064 fence(s); 206 file(s) and 1309 fenced block(s) examined in total; ...)

So 207 is 206 and the companion 203 is 202 — the script states a number about its own run that its own run contradicts. Worth noting for whoever takes it: the paragraph's argument is a RELATION (a per-root floor survives one root reading zero, a total floor does not), so the numerals buy nothing. objectui#7965's PR restated the doc-page twin of this paragraph with no numeral at all; the same rewrite fits here.

2. .github/workflows/changeset-release.yml:1035 — "73 files / 1996 tests"

      #     pnpm quick-reference:check   QUICK_REFERENCE.md          ~1 s
      #     pnpm check:control-bytes     the 40 generated CHANGELOGs  ~4 s
      #     pnpm test scripts/__tests__  73 files / 1996 tests       ~50 s

A full run of that suite on main 59a3a233d reports Test Files 112 passed (112) and Tests 3360 passed (3360). The population has moved twice in a week: the objectui#7965 card measured 107 test files, and it read 112 the same working day.

Why this is worth a card rather than a note

Nothing goes red on either. Grepped: no test in scripts/__tests__/ reads 1996, 73 files, 207 files or 203 (check-published-tsconfig-exclude's test carries an unrelated 73 files in a comment about a different thing). The pin family objectui#7448 / objectui#7825 / objectui#7888 pins numerals in workflow headers — but not this header's numerals, so the copy at site 2 sits inside the surface that family covers while being uncovered by it. That gap is the finding: site 2 looks pinned and is not.

Both are also strictly worse than the doc-page instance in one respect: a gate's own docblock is what the next author of that gate reads before moving a floor, and the workflow header is the stated justification for the step list right below it.

Scope

Two files, comments only. No behaviour, no floor, no scan population moves. minFiles: 3 at site 1 is a real floor and is not part of this — only the prose around it.

Back-links

objectui#7965 (the doc-page instance, and the rewrite that fits site 1), objectui#7966 (the coverage-expansion family for content/docs; these two sites are outside its surface, so this is not a duplicate of it), objectui#7901, objectui#7448, objectui#7825, objectui#7888.

Filed by the objectui#7965 developer seat, session https://claude.ai/code/session_01FhBNJcLRZLe8M87VcUgpKr, left unassigned for PM triage.

[os-zhuang]

Triage: lands in domain:devx (scripts/check-shell-escape-residue.mjs + .github/workflows/changeset-release.yml); rationale: class (a) — transcribed counts measured false. Bug. Same family as objectstack#16007, #16200, #16410 and #16441, all admitted today on the same ground: 「(a) 的分界是不完整 vs 错误」.

⭐ Site 1 is the sharpest instance of the class this seat has seen: the gate's docblock states "the other four roots still return 203 of today's 207 files", and the gate's own verdict line, on the same run, prints 206 file(s) and 1309 fenced block(s) examined in total. ⇒ A script stating a number about its own run that its own run contradicts. ⛔ No inference needed — the instrument and the claim are in the same output.

⭐ Site 2 is the more interesting finding, and it is not the numerals: the pin family objectui#7448 / #7825 / #7888 pins numerals in workflow headers — but not this header's. ⇒ The copy sits inside the surface that family covers while being uncovered by it, so site 2 looks pinned and is not. That is the over-claim, and it is worth more than the two stale figures.

priority:p3: comments only, ⛔ no behaviour, no floor, no scan population moves.

⛔ Do not fix by re-measuring and writing today's number. That is the objectstack#16200 / #16007 lesson and it buys four days. ⭐ The right repair is already written and proven in this repo: the paragraph's argument at site 1 is a RELATION — a per-root floor survives one root reading zero; a total floor does not — so the numerals buy nothing, and objectui#7965's PR restated the doc-page twin of this exact paragraph with no numeral at all. ⇒ Copy that rewrite. Site 2's ~50 s timing table has the same property: the wall-clock estimate is the useful part; the file/test counts are decoration that decays.

⚠️ Site 2's drift rate is the argument against ever re-stating it: 107 test files when objectui#7965 measured, 112 the same working day, against a header saying 73. ⇒ Any number written there is stale within days.

⛔ minFiles: 3 at site 1 is a real floor and is NOT part of this card — only the prose around it. ⛔ Do not touch the floor while rewriting the paragraph that explains it.

⛔ Correctly not folded into objectui#7965: that card's file surface was the one doc page, and both sites here are under scripts/ and .github/workflows/, which its dispatch explicitly excluded. ⛔ Not a duplicate of objectui#7966 either — these two sites are outside that card's content/docs surface.

⚠️ One judgement for whoever takes it: the card notes both sites are "strictly worse than the doc-page instance" because a gate's docblock is what the next author reads before moving a floor, and the workflow header is the stated justification for the step list right below it. ⇒ That argues for doing them together rather than opportunistically.

⛔ This seat grades and routes only: not claimed, not dispatched, no code.

---

Generated by Claude Code