Why is facility booking a useful test of agent reliability compared with simpler back-office tasks?

**Why not just test form filling?** Form filling measures transcription. Booking measures whether an agent knows when not to write. The approval says what was agreed; the system of record decides what actually happens next.

**What makes it harder than it looks?** Three things. Approval terms live partly in structured fields and partly in notes. Booking touches several records at once: the facility, the repayment schedule, security links and settlement instructions. And the final write starts servicing, so a mistake is expensive to unwind.

**Does it test restraint?** Yes. Some files should not be booked at all. A missing guarantor, an expired approval or a pricing exception without a named approver should produce a hold or a return. An agent that always completes the task will fail these cases.

**Is there a clear right answer?** Usually. Terms either match the approval and policy or they do not, and a hold either cites the real blocker or it does not. That makes automated grading practical.

**What should a buyer still check?** Whether the generated approvals, policies and exceptions are plausible for the product being tested, and whether any content is too close to a real source. Grounding and disclosure risk should be inspected in a sample before anything is used for training or evaluation.