[quality] tests/tools/ is required to be measured but scored by no source coverage gate #1192

Finding

tests/tools/ is a source root for one of the reporter's guarantees and
invisible to all the others.

const SOURCE_ROOTS = ['scripts', 'src', join('tests', 'tools')];

— tests/tools/coverage-report.mjs:91

// Coverage records exist for the test files too. They are not the code this
// repository ships, so they are summarised separately rather than folded
// into totals the suite's own near-complete self-coverage would dominate.
export function isSourceFile(file) {
  return !file.startsWith('tests/');
}

— tests/tools/coverage-report.mjs:448-452

SOURCE_ROOTS is what --require-source-files walks, so the gate insists
every module under tests/tools/ be measured — tests/coverage-report-source-files.test.mjs:95-99
spells out why: "The harness tree is enumerated too: a tool nothing imports
is the measuring apparatus going unmeasured."

isSourceFile() is what builds the src files row, and it drops everything
under tests/. So --check-source, --check-source-regions and the
per-file floor added in #1190 all score scripts/ and src/ and never
tests/tools/. The apparatus must be measured and is then held to no
standard. The only floor it falls under is --check-regions 95, an all-files
aggregate over 8708 regions that the suite's own test files dominate.

This is not symmetry for its own sake. tests/tools/ is where the coverage
reporters, the Playwright coverage fixture, the data-overlay loader and the
JSX/DOM harness live: a gap there is a gap in the instrument every other gate
reads.

Evidence and provenance

• Unit — measured. npm run test:unit:coverage (TZ=UTC node tests/tools/coverage-report.mjs, node v26.10.0) run locally on 2026-10-08
  at 03cfcfe. Every tests/tools/ row:

  file	line %	region %
  tests/tools/e2e-coverage-report.mjs	100.00	93.20
  tests/tools/e2e-coverage-run.mjs	100.00	98.65
  the other 13 modules	100.00	100.00

  The same run reports src files | 100.00 | 99.92 | 8984/8984 lines | 2545/2547 regions, which is how a 93.20% file sits in the tree while the
  gated row reads 99.92%.

• Both sub-100% harness files are at 100% lines, so the line gates are
  structurally blind to them as well: their gaps are sub-line regions
  (an unexecuted ?? fallback or ternary arm), the same shape [quality] the unit coverage gate is aggregate-only: 23 source regions can go uncovered without failing it #1190
  describes.

• E2E — not applicable. This is a property of the reporter's gate
  configuration, not of a code path, so no end-to-end coverage claim is made
  and this is a coverage-reporting finding rather than a coverage-gap.

Recommendation

Give the harness a floor of its own rather than folding it into src files
— the comment at coverage-report.mjs:448-450 is right that the suite's
self-coverage would dominate a merged total, and that reasoning does not
extend to leaving the harness ungated:

• tests/tools/coverage-report.mjs summarises tests/tools/ as its own
  row beside src files and all files, built from the same
  SOURCE_ROOTS entry --require-source-files already walks
• it accepts a floor for that row (line and region), failing with the
  file, ratio and uncovered region start lines the way the other gates do
• package.json's test:unit:coverage:check passes the floor, set from
  what the harness measures at the time it lands
• tests/coverage-gate-thresholds.test.mjs records the matching floor

Coordination — blocked on #1191

This must not start until #1191
(hold-gated, for #1190) lands or closes. That PR edits
tests/tools/coverage-report.mjs's flag table, report() return value and
gate block, package.json's scripts.test:unit:coverage:check line and
tests/coverage-gate-thresholds.test.mjs — the same four places every box
above edits. Two PRs over report() and the same gate string would conflict
on every merge.

It is disjoint from the other open hold-gated PRs. #1163 and #1174 touch
tests/tools/e2e-coverage-report.mjs and tests/tools/e2e-coverage-run.mjs
— the two files this finding measures, not the reporter that would gate
them; whatever those PRs do to their region percentages, the structural hole
is unchanged and the floor is set after they land. #1161, #1165, #1166,
#1168, #1171, #1176, #1178, #1181 and #1185 touch neither the reporter nor
the gate. #1183 touches package.json but adds an engines block, not a
scripts entry.

No workflow file is involved: .github/workflows/ci.yml:40 already runs
npm run test:unit:coverage:check.

Priority

• Impact: medium (the measuring apparatus is required to be measured and then
  scored by nothing; two of its fifteen modules already sit below every floor
  the repository applies to shipped code)
• Effort: medium

🐝 Hive Agent: quality | Instance: hosted-available-lke648397-260827-5n31 | SHA: 1e91668

— hive: agent=quality backend=copilot model=claude-opus-5 copilot=1.0.88