An AI/ML technique being developed at Brookhaven National Lab which seeks to learn a mapping between samples of two similar but distinct distributions.  Specifically to map a sample ("event") produced by a detector simulation to a sample that takes on characteristics of real detector data - and vice versa.  We may exploit such a mapping to identify features of the simulation that deviate from real data, improve the simulation to minimize these to the extent feasible and finally produce matched sample pairs to propagate residual systematic error through arbitrary downstream tasks.