I've finished generating 500k piminus events, 50k pizeros, and 50k single photons (all sitting in magic_single_particle with the earlier samples), and have put them through the same cuts as used earlier, adding in two fiducial cuts at Jan's suggestion.  Here are the results:

A quick tour:  The first two cuts are fiducial - removing the area near sector boundaries (very aggressively) to avoid problems of using two separate SMD sectors, and removing eta>1.5 where tracking no longer works.

The next two cuts require minimum E_T in the high tower and the cluster as a whole

5 is a tracking cut, requiring no tracks less than 10cm away from the weighted center of the tower cluster.

6 is a cut on postshower energy as a function of tower energy. The value of the constant was taken from a by-eye fit to the distribution.

7 and 8 are checks to make there is at least some energy in the SMD layers, so that we can be confident that the clustering there will not die.

9 and 10 are cuts on SMD energy as a function of tower energy.  The value of the constant was done as in 6.

11 is currently a sanity check that the clusters we find in the SMD layer are close to the weighted center of the tower cluster.  Thanks to our removal of sectors A and E (cut 1) we see that essentially all clusters pass this cut.  It can be tightened in the future.

12 is a cut on the fraction of the SMD energy in each plane that is not part of the cluster, using only the side with the higher energy in each plane.  In the future, I may do this by attempting to make a cluster on each side of the central cluster instead of simply summing the energy in a set window.

The first is normalized to 1, the much larger piminus smaple meaning we can believe the green line even out to the last cut.

This second is not normalized, so you can see how much the three large samples dwarf the others.  The next step would seem to be generating more electrons, since they are expected to successfully pretend to be photons if their tracks are not found.  Also note the cut on the number of preshower layers above threshold - it cuts more heavily (in fraction of total) on pizeros than photons, and the resulting ratio of one to the other gets better the harder the cut.