Turn a business rule into an executable check.
Start with what must remain true: a round trip should preserve a value, applying an operation twice should preserve its result, or one tenant should never receive another tenant’s records. Supported fast-check properties bind that expectation to explicit source targets, input generators and execution limits. Your team reviews the expectation before the run can use it.
Keep the inputs behind a failure.
Recorded seeds and counterexample paths support repeat investigation of supported generated properties. A separate finite combinatorial mode checks approved combinations of factors and retains exact case identities, expected results and missing coverage. These modes expose different evidence. Finite combinatorial replay preserves the selected cases and does not claim to find the smallest possible failing input.
- Check supported invariants against real source functions.
- Review which input combinations were executed or excluded.
- Replay an archived failure against a repaired source revision.
Choose a rule your team can explain.
An evaluation works best with a bounded rule and an agreed way to judge the result. Input limits, generator quality and the supported runtime affect what the run can establish. Budget exhaustion and missing required cases stay visible. A successful replay verifies the selected case; a broader campaign is needed to assess the rest of the approved scope.