The improvement in success rates achieved by passing ML-generated candidate materials through stability and property filters from pre-trained models (e.g., universal interatomic potentials) as a low-cost post-processing step.