When DM-32436 is implemented ScarletDeblendTask will once again return two different output catalogs: one is a set catalog for each band using the scarlet models (convolved to the observed seeing) and the other is a set of catalogs for each band that uses the scarlet models to re-weight the observed images so that flux is conserved (similar to the SDSS and old HSC deblender outputs). Both sets of catalogs are identical except for their heavy footprints (even though the spansets for each footprint are identical).

The reviewer for DM-32436 brought up the excellent point that this is redundant information, since the catalog with the scarlet models can be easily used to generate the conserved flux footprints and might not warrant doubling the output from the deblender. So I'm wondering if we should continue to just output the scarlet models and have downstream measurement tasks contain a config parameter to re-weight the deblender outputs, which they would do after the catalog has been loaded? The main counter-argument that I can imagine is that if we have to do this multiple times this might begin to become computationally expensive, so I'm curious if anyone knows how frequently we load catalogs from disk in the DRP pipeline, for example.

Can anyone think of any other arguments for or against producing a single deblender model catalog in each band and doing the re-weighting when the data is loaded (if so desired)?