Fit Bias in roofit

I am having a problem in my generation for fit bias using the PULL study.
I have two signal peak both described by Sum of two Gaussian.
Background describe by Chebyshev 2nd order polynomial.

What I am doing is :

  1. Generate one signal peak (600 events )
  2. Generate Chebyshev 2nd order polynomial (150 events)

Using the PDF which consist of two Double Gaussian and chebyshev 2nd order polynomial I fit the generated sample and foud PULL coming out to be negative for my signal which I have not generated.

I investigate the reason and found out that it is due to Chebyshev 2nd order generation. Some time there is a dip which make my signal peak to go negative and due to this I am getting negative bias.

For reference I am attaching one of the plot which are quite negative. I am not sure that this bias is due to fit or some problem in generation.

Can anyone put some more light???

P.S. Please see the attachment as here I may not be able to state my problem in a nice manner/ or correct manner.