I fixed a bug with the way I was assembling the SMD clusters - it was essentially getting only the left part of the cluster correct, and immediately dying on the right, meaning that if the higher of the twin peaks for a pion were on the left, the sided residual would come out pretty far off.

With the correction in place, I've ground out a new sequence of potential cuts:

The long set at the end is an exploration of how to use the fact that single photons are less likely to convert in the first preshower layer than pizeros.  Since this is a statistical process, it really begs for something other than a hard cut.  An LDA would let us keep the high-purity photon smaple with no lit preshower layers without sacrificing the huge numbers that convert early enough to light at least a few.

I was worried that applying this cut might be mixing the preshower behavior together with the pt distribution, so the following two plots show the final pt distribution with a weak cut on the number of preshowers, and the second a much tighter cut.