Which features make a handover worth including in an agent benchmark?

Use this checklist to select generated handover cases that require meaningful coordination and support a defensible evaluation.

- Include a vocabulary mismatch: the same status means different things to the sending and receiving teams. - Include an authority boundary: the recipient can investigate an issue but cannot approve the proposed remedy. - Include a dependency outside the main case: progress relies on work tracked by another operational function. - Include competing queue priorities: urgency must be justified from evidence rather than copied from a request. - Include context held in different records: the agent must connect a decision to the condition that limits it. - Include a legitimate return path: the recipient can reject an incomplete transfer with a specific reason. - Include a meaningful stopping point: the agent must recognise when further action would exceed its role.

A case need not contain every feature. Choose a small combination that produces a clear decision and an observable consequence. Record which capability each feature is intended to test, so difficulty comes from the business task rather than accidental ambiguity in the benchmark.