FEATURE-ci-kqueue-focused-diagnostics: Record validation before native capture

Author: etr

specs/tasks/FEATURE-ci-kqueue-focused-diagnostics.md

@@ -17,7 +17,7 @@ Question: which original poll/kqueue normalized trace entry differed in the actu
 - [x] Reuse the retained supported manual-dispatch `release.yml` pattern as reference only: sentinel input, exact branch, attempt 1, read-only permissions, no tag/push/release/deploy trigger. Normal `v3-release-acceptance.yml` unchanged.
 - [x] Restrict native setup to unchanged compiler/runtime selector, pinned provider, TLS-off installed library/headers and original owning translation unit. Match original `-O0 -g`, EUID 0, explicit static archive and original compile/link arguments; record current artifact identities and provenance differences honestly. Existing selector cancellation sanity runs once as required setup, separately from kqueue comparisons. No LLVM libc++ DYLD override or debugger.
 - [x] Offline TDD with Python fake children only: 16 tests cover full-first/focused-rest/cap, first/focused failures, mismatch, invalid/incomplete receipt, timeout/crash/launch failure, durable debit, reentry and live concurrent refusal. Syntax, source preservation, changed-file cpplint and actionlint checked locally.
-- [ ] Caller full Groundwork validation and review before any outward diagnostic push/dispatch.
+- [x] Caller full Groundwork validation sealed PASS: session `299ae836d87b66c04702c24a625d6472`, source `de048261307b93336d962ee24ae28e5e3552e085`, nine approvals, one iteration, zero fixed, one approved unworked finding retained by helper. Native capture remains pending, 0/64.
 - [ ] One native Darwin batch and honest terminal accounting, within the original approved 64-attempt cap. No Linux C++ test execution or native comparison has occurred during implementation.
 
 Primary evidence: `.debug/ci-47c-darwin-suite-receipt-3sqav3tc/journal.md`; implementation receipts under its sibling `diagnostic-implementation/`. Original commands and both original suite JSON receipts remain retained there. Plan: `.groundwork-plans/FEATURE-ci-kqueue-focused-diagnostics-plan.md` in the main checkout.

specs/unworked_review_issues/feature-ci-kqueue-focused-diagnostics_validation_299ae836d87b66c04702c24a625d6472.md

@@ -0,0 +1,11 @@
+# Unworked Review Issues
+
+**Run:** 2026-10-25 01:17:32
+**Task:** FEATURE-ci-kqueue-focused-diagnostics
+**Total:** 1 (0 critical, 1 major, 0 minor)
+
+## Major
+
+1. [ ] **test-quality-reviewer** | `scripts/test-diagnostic-kqueue-batch.py:156` | missing-test
+   The only batch-deadline test supplies batch_seconds=0, so it proves refusal before the first launch but does not exercise the required cumulative 300-second execution bound. The existing 20-second process-timeout test is independent. Regressions that stop subtracting elapsed batch time, or replace min(process_seconds, remaining) with process_seconds, would still satisfy these tests while allowing subsequent native execution beyond the approved batch budget.
+   *Recommendation:* Add focused offline tests with controlled time/process boundaries to verify that elapsed time across a completed first attempt prevents a later launch when the batch budget is exhausted, and that a launch with less remaining batch time than the process budget receives only that remaining time. Keep all children fake and execute no C++ or native program.