Record successful full run 27 and close finite C08 qualification

[KeyffMS]

Completed live qualification

Owner-initiated full run #27 completed successfully on the authenticated self-hosted runner. Executed source: a9cdcdc1e0cad70e88b60869e75e4166046dd306; Codex CLI 0.153.4; model gpt-5.6-sol; Debian 13. All C01-C16 are REPRODUCED, including the finite C08 replacement. C04 is informational; all fifteen required capabilities passed.

• C08 negative: actual missing-checkpoint PreCompact stop, expected exit 1, 10.643 seconds, no timeout.
• C08 repaired: real checkpoint, pressure -> automatic compaction -> SessionStart(compact) -> finish reconciliation -> completed positive output, exit 0, 36.265 seconds, all nine checks true, no timeout.
• C09: three reconciliations and two ordered automatic recovery cycles, exit 0 in 60.679 seconds, no timeout.
• C10: independent startup and post-compaction exact context proofs, exit 0 without timeout.
• C13: retained project-scoped known-error-gated non-ephemeral fallback; ephemeral transport is not claimed as working.

Immutable import and scope

Archive qualifications/34140846679/evidence-artifact.zip is the exact downloaded inner archive: 166 manifest-listed files plus manifest, three hidden fixtures, no missing/extra members or SHA-256 mismatch. Outer SHA-256: 9583393745fc8292de701c51064e5569f24092e978da7164c8f50070e7d35e73; inner: 83beb50d27198c2cc0ecd72bc6ca4b1ec2d6570bcb23cc0d56c12ec97408ec75.

Current C01-C16 packages are exact copies from this ONE full archive. Provenance retains the actually executed source SHA, not the import SHA. qualifications/index.json points to #27 and records C08 closure. Historical #25 and qualified product hashes remain unchanged. No partial results are spliced into old evidence.

README, qualification status, runbook, release instructions and changelog reflect the completed result. Additional integrity regressions verify the committed archive/summary and reject edits to copied capability evidence. Candidate PR checks intentionally remain usable before fresh live qualification of changed product bytes; strict production validation still rejects such changes. Existing tests reject historical C08 timeout, product mutation, archive corruption and promotion of unexecuted templates.

Verification

Reviewed head 5bc127f6029f72e3079837b40108783ce2e8d836, exact tree 5a60aaead2bf0f86b6ce0a35a054a5b4708faf00. Actual package validation and strict release_check.py (without --candidate) passed on the clean local snapshot of this exact tree. Local core: 67 passed; closure tests: 8 passed. The hosted preparation job verified the complete harness before the final evidence-test/changelog adjustments. Final protected CI #124 is being checked before merge.

CI #123 attempt 1 recorded a loopback-only discrepancy: all three commands exited 0 and the loopback peer received successful receipts, while the CLI-event observer reported the first receipt missing. It was correctly blocked, not accepted or weakened. The original log and diagnostic repeat are preserved in Actions and the PR discussion; this is not a failure of the completed live #27 archive. No product/runtime change was made in response.

No changes to .agents, .codex, tools, normal workflows, sandbox/approvals, authentication ownership, or release validators. No new self-hosted run, production tag or release. Temporary staging scripts/workflows are absent from the final tree and squash diff. Final PR and post-merge CI must pass before closure is declared complete.

Closes #7.

[KeyffMS]

Final CI #123 attempt 1 was not accepted: the unchanged C09 loopback test produced three completed commands with exit 0, two automatic compactions and a completed turn, but its structural CLI-event observer recorded c09_first.c09_receipt_ok=false. The loopback peer independently received successful first/second/finish receipts. The failure is therefore a discrepancy between observations in this hosted conformance run, not a timeout or a failed live #27 trial. The full command output is intentionally not retained, so the exact origin of that discrepancy is not established. No assertion, runtime, product, evidence or CI dependency is being changed to bypass it. I am making one unchanged-head repeat of the failed hosted checks to determine reproducibility; the original failure remains in Actions. Live #27 and its archive remain unchanged.

[KeyffMS]

Final reviewed head 5bc127f6029f72e3079837b40108783ce2e8d836 / tree 5a60aaead2bf0f86b6ce0a35a054a5b4708faf00: CI #124 (34197609191) is complete, SUCCESS, all eight jobs passed including both actual-CLI loopback probes and their protected distribution dependency. Strict python tools/release_check.py (no --candidate), package validation, and all eight closure tests also pass on the matching clean local tree. The intermittent observation discrepancy from earlier CI #123 is retained separately as #37; no runtime/assertion/security change or evidence rewriting was used to make this import pass. Proceeding through the protected squash merge; the post-merge run remains to be verified.

[KeyffMS]

Post-merge verification complete: main is c7c28cd46a07a353fdd88356c3fe6f8ff1fd6f21, exact reviewed tree 5a60aaead2bf0f86b6ce0a35a054a5b4708faf00. CI #125 completed SUCCESS, all eight jobs passed, including actual-CLI C08/C09 conformance and the protected distribution dependency. Strict production release_check.py also passed on the matching clean tree. Issue #7 is closed and its checklist now records baseline 2.3/full #27. The independently qualified source remains a9cdcdc..., correctly preserved in provenance. No further live run is needed for this unchanged-product evidence/documentation import. No signed production tag or GitHub Release was created. The earlier intermittent hosted receipt observation remains transparently tracked in #37, not suppressed or reclassified as a passing attempt.