What records, relationships and policies can be generated for conditional approvals?

A dataset for this workflow can contain the records an agent would meet around a conditional decision, generated to be consistent with each other.

**Core records.** Decisions with approver role, terms and expiry; conditions with owners, due dates and states; evidence items with arrival time and channel; and downstream work items that depend on the decision.

**Relationships.** Each condition links to one decision. Each evidence item links to one or more conditions. Work items point to the decision that gates them. Amendments link to the version they replace, so a waiver or variation can be traced.

**Policies.** Written rules that say which conditions are mandatory, who may waive them, how long an approval stays valid and what happens on lapse. Policies can be supplied as structured rules and as plain text, so you can test both rule-following and document reading.

**Exceptions.** Partial evidence, evidence for the wrong record, conditions satisfied out of order, expired approvals and conditions referred back to the approver.

**Tasks and environment state.** A task pack can pair a starting state with a goal, such as release the settlement if permitted, and a known correct outcome. A simulated environment can expose the same state through tool calls.

All of it is generated. Ask for sample output and inspect how well its structure and variety fit your process, and whether any content looks too close to something you recognise.