I investigated why there were a small fraction of events that would have
passed the selection before the skimming and not afterwards and viceversa.
The problem was due to the fact that all tracks with any electronID (even
the loosest) would have been called electrons while skimming (as opposed
to muons).

I also realized that when running on MC results are expected to be
different when running on skims or on everything, because the track
smearing is done at recoilNtp level. I verified that turning off all kinds
of smearing/killing all events in several thousands are back. This of
course can cause some  additional inefficiencies, but the skims are loose
enough that I do not think we care.

VubAnalysis rf-091602
contains the solution to the lepton mass problem.
	ciao
	ric