Copied from original trac ticket:
For calculating the 2D chisq one currently has to generate large numbers of toy experiments and histogram the generated square DP distributions. Would be better to have a routine that calculates the appropriate PDF value at the bin centre and multiplies it by the appropriate Jacobian to fill the histogram. This could then be compared with the data distribution.
More generally it could be useful to fill histograms with various quantities that form part of the likelihood function as well as the total |A|^2 * eff, e.g. amplitude magnitudes or phases.