How do I judge the quality of generated schedule data?

Quality has three layers, and each needs its own inspection.

**Structural quality.** Do the records agree with each other? Check that every foreign key resolves, that instalments sum to the contract amount, that superseded schedules point at a successor, and that dates fall on valid business days. These checks can be scripted and should pass in full.

**Behavioural quality.** Do the cases look like the work? Read a sample of cases end to end. Does the approval route match the policy? Do refusals carry sensible reasons? Do handovers include the information the next role would need? A reviewer with domain experience will spot unrealistic sequences faster than any script.

**Grounding and disclosure.** The data is generated, so ask how its patterns were informed by observed process behaviour, and what was done to avoid reproducing any real record. Inspect samples for unusual specificity, such as rare combinations of details that could point to a real case. Generation does not by itself guarantee privacy, so treat disclosure risk as something to test.

Then test usefulness. Run a baseline agent on a task sample and look at where it fails. If failures cluster on policy lookups and linked records, the pack is exercising the right skills. If every task is passed or failed uniformly, the difficulty is probably mis-set.