Why should an evaluation distinguish a valid request from an authorised change?

**Is understanding the customer's request enough?**

No. An agent can correctly interpret an instruction while lacking permission to execute it. In professional services, an administrative contact may be authorised to change appointment details but not the bank account used for refunds. The request's clarity does not resolve that distinction.

**What should the task make visible?**

A generated task could include a contact register, delegated permissions, the inbound message and a policy describing approval requirements. Contact CON-0831 can hold scheduling authority for account ACC-2704 while refund instructions require approval from the account's nominated financial contact.

**What should the evaluator look for?**

The agent should map the requested change to the relevant permission, identify any missing approval and choose an allowed next action. Asking for an approval already present is unnecessary friction; accepting a confident sender's assertion without checking the register is an authority failure.

This matters because business workflows distribute responsibilities across roles. Evaluation teams can use this distinction to measure whether agents reason about who may authorise each action, rather than treating access to a message or record as sufficient permission to change it.