Can a task pack include exceptions, distractors and cases where set-up should stop?

**Can a task pack include cases where set-up should not complete?** Yes. A pack can include tasks whose correct end state is a referral, a pending status or a rejection, alongside tasks that finish cleanly. Mixing them tests whether an agent stops when it should.

**Can it include distractor records?** A release can seed near-duplicate parties, stale approvals for the same party and similarly named products, so that retrieval mistakes have consequences.

**Are expected outcomes a single answer or several?** A task can list a required end state plus a set of acceptable variants, for instance two valid orders of creating the party and the mandate. Anything outside that set fails.

**Can the same task be generated in different domains?** The structure of approval, party, account and obligation carries across financial services, insurance, utilities, healthcare administration and professional services. Field names, rules and vocabulary change, so ask for the domain you intend to test.

**Does it include free-text material?** It can include approval notes, handover messages and customer correspondence, which the agent has to read for conditions that are not in structured fields.

**Is the output real?** No. It is generated, synthetic material. Inspect samples for grounding and disclosure risk before relying on them for any sensitive use.