What should we ask for in a first sample of a mismatch pack?

**Q: What should a first sample contain?**

Ask for a small, complete slice rather than a thin spread: a handful of cases per cause class, the record tables they depend on, the policy table, a schema file, a validator and reference trajectories. Completeness lets you run the whole loop.

**Q: Should we ask for grading material in the sample?**

Yes. Predicate definitions, acceptable and unacceptable end states and rationales are the part most worth inspecting, and they show how the cases will be scored.

**Q: What limits should we expect?**

The sample is generated and covers only what was scoped. It will not represent every variant of the workflow, and it should not be read as evidence of how any real organisation behaves. Volumes, formats and delivery arrangements are settled in scoping, not assumed.

**Q: Can we request our own edge cases?**

Yes. Describe the behaviours you want to test, such as a duplicate that spans two periods, and ask whether they can be included. Treat the answer as part of the scope.

**Q: What would make the sample a pass?**

Define this before it arrives: the validator passes, reference trajectories run in your harness, domain reviewers rate the cases plausible, and a disclosure review finds nothing that resembles a real party. Then decide on a larger scope.

**Q: Who needs to review it?**

An engineer for the loader and tool mapping, a domain reviewer for process realism, and whoever owns data governance for the disclosure check.