What checks should we run on a sample before accepting it?

Run these checks on a pilot sample before deciding whether it is fit to list.

- **Hand solve:** pick a spread of cases across causes and domains and solve each from the visible evidence alone. - **Gap arithmetic:** confirm that the difference between the records equals what the stated cause would produce. - **Policy permission:** check that every expected action is allowed by the attached policy and the acting role's permissions. - **Leakage scan:** look for answer clues in field names, record ordering, case titles and note wording. - **Stop cases:** confirm that some cases correctly require escalation or no change, and that they are labelled. - **Baseline test:** run a simple rule, such as matching on amount, and see which cases it solves; those cases may be too easy. - **Replay test:** run a case in the environment twice from reset and confirm the starting state is identical. - **Disclosure note:** record any grounding or disclosure concern and raise it before onward licensing.