What records and relationships could a generated escalation dataset contain?

A generated release for this workflow could be organised around the case and the rules that govern who may decide it.

**Rules and policy:** an authority matrix by role, product and amount band, plus policy text covering exceptions, dual approval and out-of-hours delegation.

**Cases and parties:** applications, claims or requests with the parties involved, exposures, supporting evidence items and the status history of each case.

**Actors:** a roster of reviewer and senior roles with their limits, and some with temporary delegations or absences.

**Process records:** case notes, escalation requests with reason codes, senior responses, returned-for-information events, outbound communications and audit events.

**Relationships that matter:** - each case points to the authority rule that applied at the time of review - each escalation points to a note that justifies it - each decision points to the actor and role that made it - related cases can share a party or loss event, which lets a grader detect splitting

A task pack adds a starting state, a goal, the allowed tools and the expected end state. A simulated environment adds a transition logic that updates these records when the agent acts.

All of this is generated. It is shaped by how real organisations handle escalations, but a buyer should still review grounding and disclosure risk, and should scope which domain and policy variants matter in a pilot.