What makes a schedule task worth using to test an agent?

Not every schedule task is a good test. Pick tasks that have most of the properties below, because these are the ones where agents tend to differ from one another.

- **A policy decides the answer.** The correct action depends on a rule the agent must find and apply, not on common sense. - **History matters.** Earlier deferrals, missed payments or amendments change what is permitted now. - **More than one record changes.** The schedule, a payment instruction, a reminder and a case note must all stay consistent. - **Approval is conditional.** Some requests are within the agent's authority and some must be handed to a named role. - **Refusal is a valid outcome.** At least some requests should end in a decline or escalation, with a recorded reason. - **Calendars create edge cases.** Month ends, public holidays and cut-off times shift dates in ways that are easy to miss. - **Mistakes are quiet.** The wrong result does not raise an error, so a grader has to inspect state to find it.

A task pack built around these properties is more likely to separate careful agents from ones that merely produce plausible schedules.