Policy outcomes
Every result has a reason.
The policy engine evaluates every predicate in the active mandate against a quote and returns exactly one of three typed outcomes. Unknown, missing, or malformed policy and payment inputs fail closed as STATE_UNAVAILABLE rather than falling through to a default.
| Outcome | Trigger | Effect |
|---|---|---|
authorized | No failing predicates. | A single-use capability is minted and the request can settle. |
escalation | Only spend-limit predicates fail; every other check passes. | No payment or capability is issued. The agent can propose an amendment for fresh Ledger approval via /finity escalations approve. |
refused | Any predicate other than a spend limit fails. | No payment, capability, or credential access is allowed. Nothing is retried automatically. |
Every outcome — including refused and escalation — produces a signed DecisionReceipt and an HCS DECISION commitment. There is no code path that reaches a purchase attempt without first producing this record; finity-verify reconstructs and checks it independently of the broker that issued it. See the CLI reference for how to run that check against public state.