In the post before the previous one, I showed all the soft lepton fits using 3 components.  Most of the fits did not converge.  At the time, Scott suggested that this was because the templates did not use the high luminosity MC and thus had holes in it.  I added the high lumi MC to the templates and there were still convergence problems.  The solution was found by using the data-based mistag shape for the light template rather than W+jets MC.  The corresponding fits are below.  Note that the chisquare is not a real goodness of fit statistic because this is an event-by-event multidimensional fit.

I think I may understand the excess of events at high pTrel.  I will show the explanation in the next post.