When an AI system changes, what evidence must exist before that change is trusted?

Git tells you what changed. HodgeForm compiles the authority in the exact artifact into a frozen gate, admits only typed evidence, and emits a signed RELEASE or BLOCK receipt that CI can verify without trusting this dashboard.

Models propose · Evidence establishes · Policy decides

Enter the gate Verify a receipt

exact artifact

→ semantic capability diff

→ frozen gate policy

→ admissible evidence

→ human approval

→ signed RELEASE / BLOCK receipt

What changed?

Freeze the exact artifact, then surface the semantic authority delta—not merely a text diff.

What evidence counts?

Each frozen obligation accepts only typed evidence at the required independence. An LLM pass cannot substitute for missing proof.

Who approved it?

The human authorization boundary is recorded with the exact gate. CI independently verifies the signed receipt against a pinned key.

A HodgeForm receipt records satisfaction of a declared gate at a declared scope. It is not a universal certificate of safety, compliance, or correctness.