Current full coverage merge exhausts 8GiB after 1337 passing tests, blocking all release gates #177

Current hosted failure, not a runtime assertion defect

At PR #174 head 30b94f158815c9e2f8eac765ba7a4aafbb017f28 (package metadata 2026.10.24), all four npm/Bun release jobs in run 40101464708 terminate with V8 heap exhaustion. This is current-head evidence, not the superseded earlier PR checks.

Evidence and reproduction

• Node22 npm job: npm run release:check; 1337 tests, 1337 pass, zero failures/cancellations/skips, then FATAL ERROR: Reached heap limit Allocation failed - JavaScript heap out of memory during JSON parsing after the test summary. Heap approximately 8066 MiB; job exit 1.
• Node22 Bun release job: same 1337/1337 passing summary followed by heap exhaustion.
• Node26 npm and Node26 Bun also report fatal heap exhaustion.
• Exact coverage runner launches c8 with --max-old-space-size=8192, collecting all test/child receipts. Its comment documents the earlier ~4GiB failure and heap increase. The current 8GiB ceiling is now exceeded.
• Ordinary partial-coverage error text earlier in these logs belongs to the deliberate coverage-gate self-test and is followed by that test passing; it should not be misclassified as the terminal failure.

Expected: the required full-source coverage/release gate finishes on the supported CI matrix with complete exact receipts. Actual: passing test execution cannot produce the final coverage/release acceptance, blocking current PR verification.

Goal and impact

The current canonical closeout item explicitly leaves final npm/Bun CI acceptance pending. The strict runner requires complete authored-source inventory and exact 100/100/100/100, with no exemptions. Preserve those guarantees while bounding coverage aggregation cost.

Limits

Hosted reproduction only; no expensive duplicate full-suite/OOM run was launched. Independent bounded source compatibility checks on this head pass 86 selection/attribution tests, two mocked workflow publication controls, typecheck and build using SDK2026.10.24 rather than the project pin2026.10.20. These do not certify the pinned release or the full PR. The specific allocation source is not independently profiled; shard/streamed aggregation is an opportunity, not a proven fix. No npm publication or deployment is inferred.

Acceptance / regression criteria

• Full npm and Bun release gates finish on both supported Node22/26 legs at the final head.
• Retain all authored modules, child-process evidence, exact thresholds and semantic assertions; no source exclusion or weaker safety to obtain a green result.
• Record peak/resource evidence and show aggregation remains within the supported runner budget; exercise representative expanded workflow/child-process receipts.
• Preserve fail-closed behavior for missing/stale/incomplete coverage receipts and publish the verified CI links.

Dedup: all 176 all-status issues/PRs inspected, plus current closeout tracker and coverage runner; no existing 8GiB-exhaustion owner found. This finding is distinct from issue175's repeat-import history defect.