The current deblender has problems in the cores of galaxy clusters in HSC data, but the same algorithm worked fine for SDSS. We believe (but should check) that this is because the HSC data is going deeper than SDSS, and faint fuzzies in the outskirts of the brighter galaxies are causing problems. If that is the case, it might help to deblend hierarchically by first deblending the bright galaxies (e.g., brighter than SDSS level, ~ 22.5 mag), removing those and then deblending everything else.
There are some rules to making changes with the deblender: given a set of peaks, the deblender should produce one output source for each and every peak. In order to satisfy the multiband processing scheme, peaks may not be dropped (for consistency across bands).