Query validation
Before reporting a metric, prove that the query answers the intended question and survives the data’s awkward cases.
Define the metric
Document population, grain, timeframe and status rules. “Revenue” may mean booked, paid, net of refunds, or another measure. The SQL cannot resolve an undefined business term for you.
Reconcile independent totals
Compare the output with a simple baseline. Check COUNT(*), distinct IDs and totals separately. A complex query matching itself is not independent validation.
Test edge cases
Include no activity, NULLs, duplicate keys, equal values, boundary dates and unmatched joins. Small samples designed to expose an error are often more useful than another large random sample.
Assertions and assumptions
A validation query can return only exceptions: duplicate keys, mismatched totals or invalid dates. An empty result is meaningful only when the check itself is correct and the tested population is present.