Hi all,

 most of the causes of the disagreements data-MC and MC-MC are
now better understood.

 I updated the page:

  http://www.slac.stanford.edu/~daniele/vub/complots/comp.html

 and I included:

 * genericMC-cocktailMC comparison in many configurations

 * data-genericMC with different breco cuts


 The coclusion are pretty exciting.



 GENERIC-COCKTAIL comparison:

 1) using SP4 cocktail (genericMC is mainly SP4) the comparison improves

 2) a cut on the integrated purity (for instance at 60%) improves the
comparison both for the shape and for the efficiency. Look for instance at
qtot distributions.

 3) differences in the qtot are due to small difference in the total
number of charged tracks. If we take only Bch the differences are gone.


 I can say that generic and cocktail ARE THE SAME but

 - SP3 and SP4 are different
 - different breco purities imply different efficiency
 - the fraction of Bch has to be the same



 DATA-GENERIC comparison:

 1) better comparison if we cut on the purity (both integrated and single
mode purities) especially for the efficiency as in generic-cocktail
comparison

 2) break down in B0-Bch helps

 3) B0's look perfect!!!! Bch's have good Mxhadfit but very bad mnu**2,
Qtot and Mxhadfit!!!!!

 4) mnu**2 for Bch and B0 are completely different on MC. On data the
distributions seem to be pretty similar.



 Final conlusions and action items:
 ----------------------------------

 Cocktail-generic comparison problem seems to be gone but we need SIMILAR
PURITY and ratio N(Bch)/N(B0)

 Then, in principle, WE CANNOT USE COCKTAILMC since it does not contain
low purity modes (especially bias in the efficiency)

 B0's are ok on data-genericMC comparison! Strange Bch distributions on
the MC.


 Then:

 a) we should use generic MC to model the background with a cut on the
purity. I propose a cut on the single mode purity (>10%) to remove the
superblock 4 present in genric MC. The use of the cocktail imply a
correction in order to take care of the low purity modes not present in
it.

 b) we should perform a careful comparison of the Breco modes between data
and generic MC and weight properly. The Luigi's work goes in this
direction.

 c) since the statistic of the generic MC is similar to the data one we
will introduces a large error due to MC statistics. S/B is 1/2 and then
the additional error will be 1.4 times the original one. With a cut Mx<1.5
GeV the ratio S/B goes down ot ~1/1. We should consider this change in the
Mx cut.

 d) we have to understand why some variables for Bch are so different.
Actually the kinematic fit seems to correct part of the disagreement.


 Daniele