In DM-8796, it was found that:

ZOGY seems to have better performance than A&L(decorr) (better true-positive rate, similar rate false positives). Part of this may be the windowing issue (losing detections in convolution window around image edge) but I tried to prevent any transients from being placed near edges and still seeing this effect a little (~10% difference).
– Tested - could be misalignment of PSF. TRY: re-run with bigger matching radius (this is not the problem)
– Tested - could be issue with ALdecorr PSF. TRY: compare ALdecorr PSF with ZOGY psf (PSFs are slightly different)
– Tested - different PSFs are related to imperfect PSF matching. Tried other parameters for basis (e.g. make them somewhat smaller) which improved the AL result slightly.
– TODO: investigate basis function generation for A&L (and use comparison with ZOGY as a method for evaluation).