I broke down and made an ntuple yesterday, which vastly speeds up the cut-testing process.  The following sequence of five histograms shows the tower energy spectrum (summed over the cluster) for each generation of signal and background, with a sequence of fairly-naive cuts applied one by one.  The numbers in the legend are the number of clusters, out of 5000 events (bearing in mind that we expect ~1400 of the thrown particles to miss the endcap entirely), that survived the listed cut.  There is a chance that multiple clusters from a single event could independently pass, though this contribution should be smaller than the previous plots as the cluster threshold was changed from 0.5 to 2.0GeV for this set. 

These are all one-dimensional cuts, and were chosen with little forethought - though they were adjusted in value somewhat to make the best cut in that immediate area.  For a more serious attempt showing the motivation behind each cut, please see later blog posts.

Jan was quick to point out that the piminus and piplus populations are not the same, which would not be expected of particles that differe essentially only in their charge.  I looked into the code and discovered that most of the hadron samples were crashing in my code -- the culprit was a hard-coded size of a cluster array.  (For diagnostic reasons I process all of them).  By doubling the number of raw clusters possible, and adding in a check that I never try to fill more than 40 (the hardcoded number), the jobs now complete.  Newer plots will show that piplus and piminus are now in good agreement.

These plots are now very easy and quick to make, so more are likely forthcoming.