What should I list when specifying a simulated escalation environment?

Use this list when writing a specification for a simulated escalation environment, so the first sample matches what your evaluation needs.

- Reviewer roles and tiers, each with a named next approver and an absent-reviewer rule. - Authority matrix rows by product, risk grade and amount, with policy references. - Applicant and related-entity records with facilities, balances and relationship types. - Tool actions the agent may call, marking which ones change case state. - Seed cases covering within-authority, over-authority, connected-exposure and ambiguous situations. - Senior-reviewer behaviour: fixed rules, scripted responses, or a simulated reviewer that can return a case for information. - Reset rules that return every record to a known state between runs. - Logging fields that graders will need, such as rationale text, cited clause and evidence IDs.

Not every item is required for a first pack. The first four are the minimum to make routing testable; the rest add realism and grading depth.

Mark each item as required, preferred or out of scope, and share the list with MirrorFirm before the sample is generated. Whether an item is included, and how deeply, is confirmed in the pilot scope.