[hivecommons-hive]

Test Improvement

splitRangeAtBoundaries (tests/tools/e2e-coverage-report.mjs) is the cut that keeps a
V8 range inside the module it started in. v8-to-istanbul abandons any range whose start
and end map to different originals, so "the split never returns the unsplit range" is a
postcondition the rest of convertScript leans on — and nothing asserted it.

The function's last line cannot take its own fallback arm:

  if (range.endOffset > start) {
    parts.push({ ...range, startOffset: start, endOffset: range.endOffset });
  }
  return parts.length === 0 ? [range] : parts;

cuts is built under offset > range.startOffset && offset < range.endOffset, and the
cuts.length === 0 guard returns early, so on this path start is the last cut and sits
strictly below range.endOffset. The tail part is always pushed; parts.length === 0 is
never true. An edit that made it true would return the straddling range, v8-to-istanbul
would drop it, and the two specs that exercise this cut today
(tests/e2e-coverage-source-boundary.test.mjs,
tests/e2e-coverage-source-boundary-segments.test.mjs) would stay green — both drive a
whole fixture bundle and assert on a rendered region count several transformations
downstream, which one dropped range need not move.

This PR therefore:

• exports splitRangeAtBoundaries beside the existing
  getLineCoverage / getRegionCoverage / isPhantomRegion / isContradictedRegion
  block, so postconditions can be stated on the offsets rather than on a report;
• adds tests/e2e-coverage-split-range.test.mjs — the no-cut identity, the strict
  endpoint comparisons (a boundary on an endpoint must not cut), the deliberate
  cut - 1 gap and why it exists, property preservation through the ...range spread,
  the tiling (ascending, non-overlapping, ending at range.endOffset), and —
  exhaustively over every cut list of a 7-offset space — the invariant that a
  non-empty cut list always leaves at least the tail part. The exhaustive test also
  asserts its own non-vacuity, so it cannot pass by never reaching the cut path;
• records at the fallback that it is unreachable, why, and which test holds it that way,
  so the zero region reads as a documented residual rather than a missing fixture. That
  is the form 4184f33 settled on for the e2e side; deleting the arm would recover one
  region but delete the statement of the invariant from the code.

Tests only. No production code and no workflow file is touched.

Verification

At 4184f33, node v26.10.0, TZ=UTC:

• node --test tests/e2e-coverage-split-range.test.mjs → 8 tests, 0 failures.
• Non-vacuity, against the real function (mutated, asserted, reverted): weakening the
  tail guard to range.endOffset > start + 100 fails 5 of the 8, including
  every cut list leaves at least the tail part. Before this PR, that mutation leaves the
  entire suite green.
• npm run test:unit:coverage:check (full gate flags) → exit 0;
  src files | 100.00 | 100.00 | 9135/9135 lines | 2753/2753 regions unchanged,
  harness files | 100.00 | 99.33 | 3921/3921 lines | 1473/1483 regions.
• npx prettier --check clean on both files; npm run check:spelling → 0 issues.

Related Issue

Refs #1232 — this lands the folded finding carried on that issue
(<!-- hive-finding-folded -->: splitRangeAtBoundaries' empty-parts fallback is
unreachable and its postcondition is unasserted) and nothing else. #1232's own Half 2 —
adding --check-source-file-regions to the e2e-coverage render step in
.github/workflows/ci.yml — is deliberately not in this PR and remains open
(needs-human: the change is inside .github/workflows/, which this agent's token tier
cannot push; GitHub rejects it server-side for want of the workflows permission, so it
needs a human or a merge-tier agent to land, and the floor it sets has to be measured
against a real browser run first).

---

Filed by quality agent (hold-gated mode). Human review required.

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

[hivecommons-hive]

Important

Held for human review by the hive's ACMM level gate.

This PR was opened by the "quality" agent while Hive policy required a human checkpoint for that agent. Non-outreach agents are held at ACMM L3–L5; the outreach agent is always held because it publishes project-facing communication.

Hive will keep the hold label until a human removes it. Operators can make a deliberate one-off release during an ACMM level change with release_level_holds=true, but level changes never release this hold automatically.