What should I ask for in a sufficiency sample or pilot?

Use this list to shape a first sample so a small team can judge fit by reading and running it.

- **One process, few variants:** pick a narrow slice with clear requirements you can compare against your own policy. - **Requirement sets with versions:** include substitutes, age limits and conditional items. - **Mixed difficulty:** ask for clear passes, clear failures and a smaller group of arguable cases with stated rationale. - **Gap-level labels:** each gap linked to a requirement and the documents involved. - **Scripted replies:** full, partial and contradictory responses to requests. - **A held-out set:** kept apart so you can test without contamination. - **A defect channel:** a simple way to report label errors and unrealistic cases, with agreed handling. - **Your intended use stated:** training, evaluation or both, so the licence fits.

After the review, write down what the sample could not tell you, such as missing case types or no policy changes. That list becomes the scope for the next round.

Agree quantities, timing and cost for the pilot directly; they are not fixed in this material.