Dstar peak in the WS D** mass difference

Let's follow the analysis naming conventions for the pions: pi* is what is being attached to a D0 candidate to form a D* and pi** is what is being attached to the D* candidate to form a D** candidate.
What we observe is what is physicslly a D* faking a D** and passing our selection in the signal WS histogram, but not in the sideband WS. This means that:

a good D0 candidate has been found
a pi* candidate has been attached to it
the D* candidate thus formed somehow acts as a D0 proxy for the D** candidate. Either:
- The pi* candidate is perturbating very little the four momentum of the D0 candidate
- The physical D0 we are dealing with was in fact partly reconstructed in the D0 candidate, and the new track being added as a pi* candidate is in fact part of the D0 decay
a track which is physically the soft pion coming from the D* decay is then attached as a pi** candidate, inducing the typical deltam* distribution in our deltam** histogram

Puzzling questions:

Why this shows up only in what the analysis uses as the signal region?
Remember that the signal region is defined on the pi* candidate attached to the D0 candidate, requiring its deltam to be compatible with a fully reconstructed D*. If the pi* candidate is fitting the explanations given above, then the yield of fake D** seen in the signal region histogram should not be much different from the one in the sidebands region!
Why we don't see the same features in our version of those histograms?
What is the D0 decay mode contributing to this leak?
Anything we can think of does not seem te quite match the D0 and D*-D0 mass cuts we are applying in the analysis!

Alessandro Cerri, Last modified: Thu Jul 22 17:30:20 CDT 2004