Which evaluation hooks make simulated handover actions inspectable?

A simulated environment can expose actions such as `inspect_case`, `request_dependency`, `propose_transfer` and `accept_transfer`. Each action result can include `action_id`, `actor_role`, `state_version_before`, `state_version_after` and a structured outcome code.

Evaluation hooks should inspect state changes as well as generated language. An agent might write that a transfer succeeded while the environment rejected its action. Conversely, the agent might reach a valid terminal state through a different permitted sequence from the reference trace.

Useful invariants include prohibiting acceptance with unresolved mandatory dependencies and preventing transferred items from remaining active in incompatible queues. A trace evaluator can also check whether repeated calls create duplicate transfers. These checks assess operational effects rather than exact wording.

For repeatable runs, a reset operation can restore a named fixture and return its initial state checksum. Record the fixture version and evaluator version with every result. A reseller can distribute the action schema and hook definitions as part of the evaluation interface, allowing downstream customers to understand which behaviours are measured and which actions remain outside the simulation.