FoCall is still using sharpness to quantify the images (I'm running the latest Mac beta). Where their 'modeling' comes in is to reduce the number of data points this software requires to make a call on the AFMA value. This is pretty evident when I load in >80 data points and look at the curve fit. Sometimes the fit is 'fair' or even 'poor' when the same data loaded into a bona fide data analysis program yields a much better fit, and a resulting value which matches what I see from visual inspection of the plotted data, but is sometimes off by 1-2 units from FoCal's selected value. Basically, I think FoCal tries to fit a few data points to a pre-specified curve, rather than fitting a curve to the data.
RE what the data points actually are, I have done what FoCal does (measure sharpness and peak contrast) with a MATLAB script and gotten the same rank-ordering of image sets. FoCal is just easier to use, as I don't need to batch-convert to TIFs, etc.
I guess the key point is that there are LOTS of wrong ways to do an AFMA, and a few right ways. As long as you find a right way that works for you, and gives you sharp images, you're fine.