In general the relative error increases of about 30%. The fit has a nice
behaviour even in the generic MC that has a pretty different signal MC
(due to different MC).
As a general comment on data I can say that looking at the fits on
data (both enriched and depleted but especially in the depleted)
you can notice that the results in the categories with 0 neutrals (ch*ne1)
are not bad while the distributions with one or more neutrals (ch*ne2) are
pretty different from MC. Then the problem should be in the neutral
reconstruction.