Urs and Henning reported some differences in the histograms produced on
skimmed MC running with and without requiring the SemiExcl tag bit.
The main observation is that the number of events which enter the ntuple
before any selection is almost twice as much when the filter is turned
off.
Looking at the per mode purity distribution of this effect
http://www.slac.stanford.edu/~rfaccini/phys/vub/purcomp.eps
(dots are 'filter on', histo are 'filter off' data)
you can see that the additional events are all in the very low purity
modes.
Running the whole analysis chain on the 10K events produced by Henning,
one notices that just requiring the presence of a >1 GeV lepton the number
of selected events turns out very similar:
(top: filter on, middle: filter off, bottom: one on top of the other)
http://www.slac.stanford.edu/~rfaccini/phys/vub/filterCheck.eps
(the number of events is 27 versus 29 ...)
The fact that the events are not the same is natural because we are
applying PID killing and therefore the two distributions can be the same
only on a statistical basis.
I am getting files (courtesy of Fabio Bellini) without PID killing in
order to see that all 'filtered' events are contained in the no-filter
sample. It is nonetheless disturbing
that I could not find a clear reason for the additional, low purity events
before the requirement on the lepton.
ideas are welcome
ciao
ric
|