What acceptance checks should I run on a sample before wider use?

Run these checks on a sample before committing to wider use; each one can be done by a script or a short manual review.

- Every allocation points to an existing bank line and open item. - Open amounts reconcile to original amounts less applied amounts and credits. - Each task has a reference outcome and the policy rule it depends on. - Reviewers who work without the reference reach the same outcome on most sampled tasks, and disagreements are logged. - Difficulty is mixed, with routine matches, exceptions and escalation cases all present. - Near-miss distractors exist, such as similar references and names. - No string in the free text looks like a real payer, account or address. - The licence terms allow the evaluation use you plan, including publishing aggregate results if you intend to.

Keep a record of which checks failed and what was changed. If several fail, return to the provider with the specific task ids rather than a general complaint, so the generation can be corrected at the source. Repeat the same checks on each new version so you can see whether fixes held and whether anything regressed between releases.