Here’s what I have so far in testing the soft electron fakes.  To make these plots, I only took events which had no tight electrons and no SecVtx b-tags.  I then ran a conversion-removal filter, and only looked at tracks which did not have a conversion partner in the event.  (in the muon sample, I also ignored any track that was associated with a muon object.)  The plots are made with a likelihood cut of 0.99.  It seems that the fake rate is quite good in the muon sample, but there are some strange effects in the jet sample:

First of all, the fake rate is higher in the jet sample than in the muon sample.  Also, there are two strange features in the Pt dependence of the fake rate.  There’s a bump at 4 GeV, which I think has to be from real electrons that have avoided my cuts.  Could there still be enough heavy flavor (after the SecVtx cut) to cause that?  The second feature is the fact that the fake rate increases at higher Pt in the jet sample, and I’m not sure what could be causing that.

Also, I think I should have used one of these samples to train the likelihood, since the training sample that I used (electron triggered events with conversions removed) seems to have electron contamination in the fake sample; the fake rate is much higher than in either of the previous samples.  Since these plots are where the parameterization comes from, I expect that this would lead to too many soft electrons in the MC.

-Scott