Grounding and authorization

The host supplies the user request, candidate records, trusted identity, tenant or account scope, and the policy context. Core records and evaluates the relationship among those inputs; it does not decide business meaning from an untrusted model response.

Grounding before authorization

Grounding narrows a proposal to an exact referent. The host should obtain candidates from an authoritative source, preserve the source revision and scope, and ask for clarification when more than one candidate remains. A model may help interpret language, but it must not silently choose a candidate or manufacture a referent.

Authorization evaluates the grounded operation against requester identity, tenant scope, policy authority, and the relevant policy revision. The host owns the policy source and trusted-context construction. Core records the decision provenance and rejects malformed, stale, or insufficient context according to the approved contract.

Dispatch boundary

Do not dispatch until the proposal is grounded and authorization permits that exact operation. A clarification result is a safe pre-dispatch outcome. See the Core API Reference for exact declarations and Core lifecycle concepts for the conceptual model.