What kinds of decision evidence could a generated handover dataset contain?

A handover dataset can contain more than case records and messages. It can include the evidence that explains why one statement should control the next action while another should remain background context. That distinction is useful when evaluating agents that must reconcile apparently consistent but differently authoritative records.

In professional services, a project transfer might contain an approved scope, an internal delivery estimate, a client's requested addition and a draft change proposal. All four can describe the same work, but only the approved scope currently authorises delivery. Treating the draft as an instruction creates an incorrect commitment.

A generated release could encode document status, issuing role, approval basis and supersession links. Related tasks could ask the agent to identify the governing decision, explain why another record does not override it and transfer the unresolved request separately.

The dataset should preserve these distinctions in both readable text and structured metadata. Evaluation teams can then vary the apparent confidence of a message without changing its authority, testing whether an agent follows the strongest wording or the actual decision structure.