These results show histograms calculated for two metrics suggested by John Conway.
1) The histogram on the left shows what has been discussed thus far as the Conway metric- a series of rms values are calculated for pixels in the difference image which in the corresponding smoothed model image fall into a range of values. The ranges of values form a consecutive set of bins, with the rightmost bin corresponding to pixels in the smoothed model whose values fall between Imax (the max value in the smoothed image) and Imax/sqrt(10), the bin to the left of this covering pixels whose value in the smoothed image fall between Imax/sqrt(10) and Imax/10, continuing along in factors of 1/sqrt(10) til we get to Imax/1000. A final bin (with xvalue on the histogram -3.5 to -3) mops up the pixels correponding to those in the smoothed with a value below Imax/1000. The rms values calculated are scaled by 1/Imax before plotting against the y axis so that the leftmost bins give a value which may be interpreted as the reciprocal of off source dynamic range. But in general, it's the scaled rms in the difference image as a function of smoothed model intensity.
2) The histogram on the right shares the same x axis as that on the left, but takes the rms values calculated and uses them to divide the arithmetic mean of the intensity range of the bin in the smoothed model image (which defined the pixels used in calculating the rms). So the y axis here is the (binned) dynamic range as a function of smoothed model intensity.
Rereading this I see that I have defined dynamic range differently for each graph- I'm sure John can do a more subtle interpretation of these measures than me. In any case they seem to complement the existing fidelity measures nicely by giving some indication of on and off source rms and dynamic range.