I'm processing a bunch of data I collected during a run of clear dark skies last week, and experimenting with the Subframe Selector tool.
The documentation suggests adding a constant to the weight to avoid negative weights. This is actually required. Image Integration throws an error and quits if any weight is negative.
Furthermore, the choice of offset in the Weight Expression is very important to the result. I'm using the following weight expression:
(FWHMSigma/-3) + (EccentricitySigma/-2) + SNRWeightSigma/4 + 1
Here I'm downweighting FWHM and Eccentricity but upweighting SNR, normalizing each stat by its standard deviation (Sigma), and adding 2 to the final result to avoid negative weights.
The values of my SSWEIGHT FITS keyword in this case ranged from 0.5 to 2.5, meaning my best sub "counts" as much as 5 of my worst sub in the integrated image.
Had I chosen an offset of +5 instead of +1 in the SSWEIGHT formula my best sub would have only counted as much as 1.4 of my worst subs.
Choose those offsets carefully!