During development, synthetic samples should be used instead of real patient data to validate the tools, and both successful and uncertain results should be checked, comparing reported counts against the generated audit rows.