A Hamiltonian-derived proof state attached to each canonical Vogon epoch JSON. Decomposed into 7 penalty components (dynamic · topological · semantic · uncertainty · economic · proof · observability), then anchored by sha256 hash to the DQLDB. Every score is reproducible bit-for-bit from the source canonical bytes.

Each record carries a small mathematical fingerprint of its own truth. The fingerprint is reproducible from the source bytes, signed, and locked into a permanent ledger. You can audit it. You can prove it.

The pipeline is monotonic. Later stages can only ADD constraints; they cannot weaken earlier verdicts. ANCHORED records are write-locked by their sha256 anchor; any change is itself a new anchored event. Schema versions are strictly additive. Trust rungs only climb (UNATTESTED → WITNESS_VERIFIED → QUORUM_VERIFIED → MERKLE_VERIFIED → SOVEREIGN_VERIFIED). No silent demotion. No silent rewrite.

Once something reaches a trust rung here, the rung cannot quietly demote. Each stage tightens constraints. Nothing reverts silently. If a record's status changes, the change is itself a recorded, anchored event.

AI on the open web has no per-record provenance. Here every record is hash-addressable canonical JSON, carries an explicit trust ladder rung, a published 7-component Hamiltonian, a verifiable DQLDB anchor, and (for anomalous records) a B.2.2 connected_but_distant explanation. AI agents can verify origin per record, weight retrieval by attestation level, and detect tampering before training or inference.

AI today reads the open internet and has to guess what is true. Here, AI reads a ledger where every fact carries a receipt. It can finally tell the difference between something it should rely on and something it shouldn't, at the record level, not by source reputation.