Fig. 21.

Various classifier metrics for a classifier trained to discriminate a combination of FETA, CATHODE, and CURTAINs synthetic SM samples from data with $n_{\text{sig}}=750$. Errorbands show a 68-percentile spread across the median and come from training a fivefold classifier 100 times with different random seeds, over 3 independent generative model seeds, ensembling scores over 10 runs, and averaging classifier metrics over the ensembles. This is a companion plot to Fig. 14