COUNCIL OF AI / MEOK / RELEASE ACCEPTANCE
Check the evidence.
Then claim the result.
A public replay defect is corrected: a zero-test replacement can no longer be labelled as 18 passing controls. This update also reconciles the live receiver, published research and next qualification steps.
Observed 11 November 2026. Scoped research acceptance, not whole-system certification or a completed human-authority journey.
The correction
The old runner accepted an altered local test file, ran zero tests, and still printed that 18 controls passed. The corrected runner verifies every executed source, captures its bytes, and requires the actual 18-test result. The deliberate substitution is now refused.
This finding does not invalidate the earlier genuine 18-test evidence. It closes a way a later altered reproduction could report success incorrectly.
Read the exact before/after cases · Download the corrected reproduction script · Inspect the 26 regression tests
Mobile results stay in view
The original populated results widened the page on a 390-pixel screen. The current owner published equivalent CSS containment and stale-result protection before this upload. Our duplicate CSS was withdrawn; their root navigation and model/data files are preserved. All 24 earlier staged checks passed; final served checks qualify the combined release separately.
Current release map
| Area | Established in this pass | Boundary |
|---|---|---|
| Private receiver | 13 live calls, seven expected refusals; event/hash/clock and recovery checks passed. | Caller pins are not authenticated by the receiver; no permission to act. |
| Public replay | Pinned runner repair and original 18-test replay. | Same-author qualification; not independent acceptance. |
| Existing model labs | Preserved source and research links. | Retained synthetic tasks, not a general-world-model ranking. |
| Saved copies | Current source records inspected. | Hosted persistence acceptance remains separate; no stored user data accessed. |
| Source release | PR #4180 ready for review; current named checks green. | Not merged into protected master. |
Read, reproduce and continue
The next decisive qualification is host-authenticated current source/context, followed by one real bounded approval and independently observed harmless effect. Training, persistence, hardware and financial claims are not implied by a successful tool check.
No new paid job, model training, source-data export, signature, human appointment, scheduler or protected merge. Historical reports retain their original dates. The published composition passed 24 desktop/mobile workflow checks, clean 26-test and 18-control reproductions, and the scoped file/link checks. Exact revisions and limits are in the dated post-publication record.