As promised, here is a first-pass set of cuts utilizing track information, and attempting to make more use of the SMD than the previous suite of cuts.

(This is for the endcap only, of course.  I really should have noted 'endcap' in the previous blog post as well.)

The basic strategy, same as last time, is to find every lit tower in the endcap and construct a cluster of four towers around it that has the higehst energy of any of the four possible clusters.  We then access information about the four postshower layers, the four (times two) preshower layers, and the smd layers covered by these towers, as well as track information from the TPC.  With this information we make a serious of cuts, listed on the first page of the attached pdf, and described in the comments that follow.

Unless otherwise noted, the windows for the following plots are always selected so that there are NO overflow events.  Additionally, the two dimensional plots are always in the same positions - gamma in the upper left, muon in the upper right, pizero in the middle right.  They have labels, but these are frequently obscured by the histogram titles.

Cut profiles count every cluster, regardless of which event it came from. 

page 1:  The fraction of reconstructed clusters passing each cut.  Requirements for reconstruction are fairly loose, so that with the exception of the muons, this represents essentially all of the clusters in each event.
page 2: The squared distance between tower clusters and the nearest primary track (in X-Y).  If no primary track is found, the distance is very large, as seen here.
page 3: A close-up of the region near zero for the squared distance between tower clsuter and primary track.  Note that the number of pizeroes in this plot is roughly twice the number of photons, which would be expected if tracks associated with these were the result of early conversion - two photons means twice the chance to have a e+e- pair form early enough to generate a track.  I'm not sure I like the number of conversions being this high, though - 10% seems like too many, and makes me suspect something else is going on as well.
page 4: the X-Y position of the closest track in each event. 
page 5: cluster X-Y after the first two cuts. 
page 6: cluster X-Y after applying the tracking cut (distance between cluster center and track must be >10cm).  The charged particles have lost a good deal of their populations at low eta.  If necessary, we can make this cut more drastic and try to remove the remaining lines.
page 7: U-plane cluster energy as a fraction fo the total energy in that plane.  (From here on we have applied a cut requiring u and v plane energies each to be > 10 MeV.  This cuts signal more drastically than background, but without this we cannot form ratios or do any other work in the SMD) Note that pizeroes have a hump extending toward a lower fraction.  This hump is likely very dependent on the exact method of clustering, and so if I adjust the clustering algorithm I will have to revisit this.  Currently, I search for the highest strip in the plane within +/-40 strips of the center of the lit towers, and then begin adding strips on either side of the seed strip until either the energy of the next strip drops below 0.2MeV, or is more than 20% larger than the previous strip.
page 8:V plane cluster energy as a fraction of the total energy in that plane.  The overall shape is the same as for the U plane.
page 9: SMD cluster energy as af raction of the total energy in that plane, for the first Z plane.
page 10: SMD cluster energy as a fraction of the total energy in that plane, for the second Z plane.
page 11: ... and for the third.  There's no significant change from plane to plane.
page 12:  The SMD cluster energy fraction in u vs v.
page 13: SMD u fraction * SMD v fraction vs tower energy.  This can be read as 'the chance for both the u and v clusters to contain a large fraction of the total plane energy'.  Since pizeroes are more likely to have split clusters, this fraction should be lower for them, and indeed this is borne out in the plot. 
page 14: the same SMD fraction * SMD fraction as the previous plot, shown now as a funciton of tower energy.  We see that the chance to have a strongly split cluster for a pizero drops at higher energy, which might be something we could fix by adjusting the clustering algorithm.  (perhaps by defining the '20%' as a function of tower energy instead of being static)
page 15:  postshower energy as a function of shower energy.  We have applied now a cut requiring SMD fraction * SMD fraction > 0.5 .  This cut is very effective at knocking out hadrons, to the point that the postshower cut no longer seems particularly profitable (in the previous iteration, we required postshowerE<.001*towerE)
page 16: Now using the postshower energy cut, this plot shows the position of the smd cluster relative to the tower cluster in XY, showing agreement to within the resolution provided by the towers.  (worse for single photons than for any other particle)  There are a few outliers (~5) that are outside of these windows, always corresponding to being ~equally far off in both x and y.
page17: This is the cutprofile shown on page one, set to logarithmic scale now so that the last few steps are more visible.  There are only two cuts that are effective against the EM backgrounds (pizero and electron) currently, #3, which is the tracking cut and chops out electrons at lower eta, and #5 which is the cluster shape cut and cuts out pizeroes at lower energies. 
page 18: a rough way to look at the discrimination for eta < 1.5, which should correspond to the area where tracking works well.  We can see that the tracking cut (now #4) is very effective at removing electrons and all hadronic backgrounds, clearing out ~80% of each.
page 19: inverting the temporary cut in the previous, we can see how the discrimination fares in the region without tracking.  The tracking cut is now only marginally effective.  The cluster shape cut, though, remains a strong discriminator between gammas and all backgrounds (save electrons).
page 20:  belatedly, I realize I set the thresholds for uE and vE much higher than I intended.  This profile shows the cut effectiveness if I reduce them to the values in the previous pdf I posted (0.0015).  This softer cut preserves more of the photon signal and also the hadronic signal, and is probably the more useful of the two cuts.

If there are other plots you would like to see, please let me know so I can generate them.