200 rows, more noise features: 20 features leaky 0.522 (min 0.415) clean 0.522 gap +0.000 100 features leaky 0.640 (min 0.595) clean 0.498 gap +0.142 1,000 features leaky 0.746 (min 0.695) clean 0.501 gap +0.245 5,000 features leaky 0.798 (min 0.745) clean 0.496 gap +0.302 20,000 features leaky 0.833 (min 0.780) clean 0.522 gap +0.312 5,000 noise features, more rows: 100 rows leaky 0.877 (min 0.820) clean 0.500 gap +0.377 200 rows leaky 0.798 (min 0.745) clean 0.496 gap +0.302 1,000 rows leaky 0.656 (min 0.627) clean 0.507 gap +0.150 5,000 rows leaky 0.570 (min 0.562) clean 0.500 gap +0.070 scikit-learn 1.9.1 | numpy 2.5.3