How do we inspect generated set-up data for grounding and disclosure risk?

**Does generated data mean there is no disclosure risk?**

No. Generated output is not a guarantee of anonymity or privacy compliance, and it is not a substitute for your own review. Treat both grounding and disclosure as questions to examine.

**How do we inspect grounding?**

Grounding asks whether the workflow behaves the way real ones do. Ask domain staff to review a sample of cases and the policy text. Look for realistic ordering of steps, plausible exceptions, and handovers that match how teams actually divide work. Note anything that reads as too tidy, or exceptions that no practitioner would recognise.

**How do we inspect disclosure risk?**

- Search the release for strings that look like real identifiers, contact details or addresses you can recognise. - Check for rare combinations of attributes that might single out a real person or organisation. - Ask what inputs informed generation and what controls were applied, and request that answer in writing. - Run your own matching tests against any sensitive datasets you are permitted to compare with.

**Who decides if a release is acceptable?**

You do, under your own governance. The pilot is the place to agree the review method and what evidence you need.

**What if we find a problem?**

Report it with the record IDs. A good pilot includes a path for removing or regenerating affected material.