… drift gate

ROADMAP.md still marked v1.24 as "(PLANNED)" though v1.24.0 (2026-06-21) and
v1.24.1 (2026-07-28) had both already shipped, and gave the locale count as
17 — stale since the 11->17 expansion (v1.24.0) plus ru/ko (landed the next
day, before v1.24.1). Same "17 locales" drift was independently present in
docs/TRANSLATION-GUIDE.md (also missing 1 of 21 modules — desktop.json — from
its module list and count), the root CONTRIBUTING.md (the one README.md
actually links to), .github/copilot-instructions.md (2 occurrences), and
docs/i18n/AUDIT_AND_IMPROVEMENT_PLAN.md's "current state" section. Fixed all
of them; left every CHANGELOG.md and dated AUDIT.md/TODO.md historical entry
untouched — they correctly describe what was true at a past release.

Added scripts/check-doc-metrics.mjs (wired into ci.yml's quality job as
`pnpm run docs:check`) so this class of drift fails CI instead of silently
accumulating again. It computes the actual locale count, key count (dedup'd
the same way check-i18n-keys.mjs does), and latest released version tag, then
scans README/ROADMAP/TODO/TRANSLATION-GUIDE/CONTRIBUTING for stale numbers or
a "vX.Y.Z ... PLANNED" claim for an already-released version.

Historical sections are exempt (Keep-a-Changelog `## [x.y.z]`, `## vX.Y.Z ...
RELEASED ...`, a heading suffixed with an emoji checkmark, a `**Status:** ✅
Released/Completed` line just below a heading, or this repo's own TODO.md "✅
= done" bullet convention) — the first version only recognized the
Keep-a-Changelog/RELEASED-heading shapes and produced 26 false positives
against TODO.md/ROADMAP.md's other historical conventions; broadened it
before wiring the gate in. tests/unit/checkDocMetrics.test.ts covers the
exclusion heuristic directly (the plan's explicit requirement — an untested
exclusion rule turns a drift gate into noise).

…n, mutation-testing status

- Re-pinned token-audit-baseline.json to the actual measured total (160,
  down from a stale 165 dated 2026-06-14) after WS-1..6 landed. Ratchets
  down only, per the project's hard rule against loosening a quality gate.
- Added scripts/check-coverage-ratchet.mjs: reads coverage/coverage-summary.json
  (added the `json-summary` reporter to vitest.config.ts — it wasn't being
  generated at all) and suggests raising a threshold when the measured value
  sits 2+ points above it, matching the "3 green runs, max 5 pts/quarter"
  rule already documented as a comment in vitest.config.ts but never
  mechanized. Wired into ci.yml as `continue-on-error: true` — informational
  only, never blocks the build; the ratchet decision stays with the
  maintainer.
- docs/CODE_QUALITY.md: added a Suppression Reduction Plan grouping all 33
  noExplicitAny suppressions (of 52 total, at an exact-match baseline with
  no headroom) into clusters by root cause and effort — no code changes,
  visibility only.
- docs/CI.md: corrected three places that still described Stryker mutation
  testing as part of the active `ci.yml` job graph (job-graph diagram, the
  `mutation` job's table row, and the CI-tier table) — it was removed from
  the merge pipeline on 2026-06-02 and now runs only via manual
  `workflow_dispatch` on mutation.yml. Documented why, how to trigger it,
  and the re-integration criterion.