FounderCLI public evidence record: exact-revision publish gate

Snapshot revision: 98ac613c18c8d10b4fa2eee57c2c64c3ecd4b9f8
Source records: docs/DECISIONS.md decisions 9 and 16, plus apps/web/tests/int/publishGate.int.spec.ts at the snapshot revision
Event date: 2026-08-17
Publisher: FounderCLI

Defect that motivated the gate

The Post editorial review status was originally a select field whose default claimed Human final review. No hook enforced the field, so a new Post could claim review without a reviewer or a bound revision.

Implemented boundary

The publish hook refuses publication while a Post awaits human review. A governor's approval records the approving user, timestamp, and a hash of the exact substantive fields that were reviewed. A later substantive edit invalidates that approval. Presentation and scheduling fields are excluded from the frozen set so changing them does not revoke a content review.

Falsification evidence

Removing the enforcement hook caused seven of the first ten gate tests to fail. Two tests that initially survived were found to assert properties of an approval without first asserting that an approval existed; after adding that precondition, nine of ten tests failed without the gate. The remaining test covered the field default rather than the hook.

Governance failure found during verification

A production cache test briefly approved and published a real article as a fixture, leaving a governor approval record that did not represent a reading. The record was cleared and the rule was changed: production verification may test refusals, but it must not manufacture a human approval. The gate enforces state and exact-revision binding; it cannot prove that the named person actually read or understood the article.