On-Chain Audit Anchor

The part nobody else has: an immutable, publicly verifiable proof of every decision — no need to trust AgentOps or the operator.

How it works

• Every decision (ALLOW or DENY) is recorded in Postgres with the full intent JSON.
• After settlement, a background reconcile worker computes SHA-256(decision_json).
• It writes that hash into the Odra GuardRegistry contract on Casper, keyed by decision_id.

Why it’s tamper-proof

Re-anchoring the same decision_id with a different hash makes the contract call revert. So a decision can be written once and never silently altered. Anyone can recompute the hash from the decision record and check it against the on-chain value on any Casper block explorer.

Legal Context Protocol (LCP)

Before spending, an agent can be shown a service’s terms (/.well-known/legal-context.json). AgentOps verifies its atrHash and attaches it to the on-chain anchor — turning the record into a legally-provable receipt that the agent agreed to specific terms.