I agree, this can also detect brittle tests (e.g, test methods/classes that only pass if executed in a particular order). But applying it for all data could be expensive computation-wise