Recent investigations in DM-30466 shows that one of the (more common?) failure modes in the astrometry fitting is erroneous matching between source and reference objects of incompatible magnitudes.

As discussed on that ticket, one of the challenges of a magnitude-based selection is that we don't have a zeropoint yet. However, we do have a match between source and reference objects. Therefore, it should be possible to compute an effective zeropoint. My suggestion is to compute -2.5*log10(src_flux) + 2.5*log10(ref_flux) and perform a configurable n-sigma cut about the mean/median when doing outlier rejection.

This will obviously require some significant testing on RC2 and DC2 data to ensure that nothing breaks.