What matters for your decision
Synthetic data does not measure real coverage, but it separates predictable architecture problems from unresolved privacy and procurement decisions. Mark results as provisional until a controlled real-data test is approved.
Run a useful comparison
- Build fixtures with dummy identifiers containing no production values, then score them by intent and expected action.
- Review the data map before moving one low-risk queue to an access-controlled pilot.