While starting to implement the magnitude based photometric model (DM-14574), I noticed that my implementation of the error model for the flux based model was incorrect. I'm going to fix that now and see if it helps the fitter.