Can a task pack include reference answers, rationales and hard negatives?

**Can a task pack include the correct answers?**

Yes. A pack could carry a reference final state for each task, such as the allocations that should be posted, the amounts that should be parked and the cases that should be opened. It could also carry an acceptable-alternatives list where policy allows more than one valid outcome.

**Can it include the reasoning behind them?**

A buyer can ask for a short rationale per task that cites the policy rule and the records used. This helps reviewers audit the reference and helps you build rubric-based scoring.

**Can I get hard negatives?**

Ask for near-miss cases: invoices with close numbers, customers with similar names, amounts that match two open items, and references that point to a cancelled document. These are the cases where weak agents tend to guess.

**Can the data include time?**

Yes. Records can carry dates, and a scenario can include later events such as a bank reversal or a late credit note. That lets you test whether an agent revisits an earlier allocation.

**Can I choose the domain?**

The workflow can be framed in financial services, insurance, utilities, logistics, healthcare administration, retail operations or professional services. The policies and vocabulary change with the domain, which is worth specifying in a pilot.

**Is any of it real?**

No. The output is generated, not real business records. Inspect samples for grounding and disclosure risk before you use them.