docs: delete three hand-maintained integers from prose, keeping the claims they propped up

[os-trump]

Fixes #1315
Fixes #1310
Fixes #1312

One defect, three instances: a counted fact stated in prose with no producer and nothing checking it. All three carry the identical ruling — delete the numeral, keep the load-bearing claim. No renumbering, and no counting guard on any of them (ADR-0049 axis 3).

Comment text only. Three non-overlapping files, one commit per member so each is independently reviewable. No assertion, import, describe block, table row or seed changed anywhere in this PR; both test files got smaller.

Verified on 2b50d55 — pnpm verify green end to end (validate, typecheck, lint, lint:i18n-gate, hygiene, hygiene:tokens, build, test). Gate verdict lines quoted per member below; suite total 142 test files passed, 2997 passed / 1 skipped.

---

Member 1 — #1315 · scripts/check-source-hygiene.mjs

scanHeaderPosition's docstring argued that requiring the licence header to be present (not merely well-placed) costs nothing, by quoting two hand-maintained integers: "all 282 .ts files under the scanned trees carry the header, 274 of them on line 1".

Re-measured on d456ea3. The card's own corrected figures were already stale, exactly as the dispatch predicted:

.ts under the SCANNED trees        : 326      (card said 324, as of 6a0e192f)
root .ts from ROOT_TEXT_FILES      : 3        objectstack.config.ts, playwright.config.ts, vitest.config.ts
allTs = the header check's surface : 329
  header at its required line      : 329
  header present but misplaced     : 0
  header absent                    : 0
control: the gate's own banner reports 339 files under SCANNED and 3 root .ts — reproduced exactly

Both integers were stale, and the eight-file gap the second one hedged against has closed to zero — so the sentence understated its own case, and left a caveat that no longer described anything.

Two things were wrong, not one. Beyond the numerals, the set had drifted: PR #1313 widened allTs to take in the root .ts files, so "under the scanned trees" had become a proper subset of what the check actually reads. The corrected sentence now names the set explicitly.

Why no count, and why this member is the strongest of the three: scanHeaderPosition is the producer. The universal is the check's own postcondition, not a hopeful generalisation — the integers were a snapshot of a state the check has since made unconditional. Saying so is both shorter and more accurate.

That claim was reverse-verified rather than asserted. Pushing the header down one line in vitest.config.ts — deliberately one of the three root files, so the widened surface is under test too:

mutation confirmed on disk: hash 7227b37 -> 77fcb11 (differs from the HEAD blob)

  gate exit on mutated tree = 1
  ✗ copyright header at the top of every .ts file — 1 violation(s)
      vitest.config.ts:2  header on line 2, must be line 1 — move it back above `'use strict';`
  ✗ source hygiene failed: copyright header at the top of every .ts file

restore: hash back to 7227b37, byte-identical to the HEAD blob, `git diff HEAD` empty
  gate exit on restored tree = 0
  ✓ copyright header at the top of every .ts file
  ✓ source hygiene clean

Acceptance: the two integers are gone; the sentence is not renumbered; no counting guard added; the set the universal ranges over is named; MAX_FILE_BYTES / ADVISORY_FRACTION / ADVISORY_BYTES and the "why 70%" comment are untouched.

Member 2 — #1310 · test/sharing-coverage.test.ts

The note above the crm_campaign row justified 「营销活动」 as "the word the other 14 zh-Hans doc pages already used".

Re-measured on d456ea3:

.zh-Hans.mdx pages under content/docs : 67   -- control: the search reached the corpus
pages containing 营销活动              : 21   -- the comment said 14
pages containing 市场活动 (pre-#830)   : 0
locale pack: src/translations/zh-CN.ts:211  crm_campaign: { label: '营销活动' }

Wrong by seven — and unlike its siblings, nothing in the file disagrees with it, so it reads as a measurement to every subsequent reader. Without the numeral the claim is stronger: a universal cannot be falsified by the next zh-Hans page someone writes, and both halves keep real producers (the locale pack is a file; "the docs already use it" is checkable by grep).

The second half of the comment is kept: the history of the partial rename in #830 has a producer, and the 市场活动: 0 reading above confirms it discharged cleanly.

Acceptance: numeral gone, not renumbered to 21; the #830 history retained; the table row, the assertions and the #810 comment below are untouched; file shrank 66,068 B to 66,058 B.

Member 3 — #1312 · test/flow-scheduled.test.ts

The suite header opened with "All six scheduled flows were previously untested at runtime".

Re-measured on d456ea3, counting entities rather than occurrences:

.ts under src/flows                    : 24   -- control: the search reached the tree
scheduled flow FILES                   : 9    -- the comment said six
textual occurrences of type: 'schedule': 10

The tenth occurrence is the string inside campaign-completion.flow.ts:11's JSDoc, above the real declaration at :19; that file exports exactly one flow (CampaignCompletionFlow). Nine files, nine flows, ten textual occurrences.

This member is the worst of the family: the file imports all nine and writes a describe block for each (10 src/flows imports, 14 describe( blocks), so the sentence undercounts the coverage of the suite it heads by three. A reader auditing scheduled-flow coverage reads "six", counts six, and stops while three tested flows sit below the fold. No assertion ever read the numeral, so the suite stayed green throughout.

No counting guard here either, and this member supplies the decisive argument: such a guard would have to solve the occurrence-versus-entity problem to be correct, and a grep-based one would pin ten — the wrong number.

Acceptance: numeral gone, not renumbered to "nine"; no guard; no assertion, import or describe block touched; file shrank 71,456 B to 71,448 B.

---

Changeset

.changeset/hand-maintained-integers-in-prose.md, empty frontmatter — the sanctioned "releases nothing" declaration this repo documents in .github/workflows/changeset-check.yml, matching the two sibling changesets already in the directory. Comment text only across all three members, so nothing ships to HotCRM users. No skip-changeset label applied — empty frontmatter is this repo's declared route.

Size advisory — reported, not acted on

flow-scheduled.test.ts sits 224 B below the 70% advisory band, so deleting words was the safe direction; it is now 232 B below (71,448 of the 71,680 band). The hygiene advisory that this run printed names only the two translation bundles, neither of them in this PR:

ℹ️  70% of the 100KB cap reached, and not passed — advisory, this does not fail the gate
    src/translations/es-ES.ts  75.3% of cap, 25KB of headroom left
    src/translations/ja-JP.ts  73.4% of cap, 27KB of headroom left

Reported rather than acted on — that band is #1311's subject, which is not addressed here.

Fences honoured

.github/workflows/code-quality.yml (#1238's file) and test/source-hygiene-*.test.ts (#1314's) are untouched. #1184 and #1311 remain open and are out of scope here. The read-coupling with #1238 was handled as directed: the surface description above is taken from the gate's own banner, not from either card's prose.

---

Generated by Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
hotcrm	Ignored		Aug 26, 2026 9:21am