In order to address issues observed in the detection efficiency in HSC coadds,
DM-12995 added a "dynamic detection" algorithm to the detection phase on coadds. Essentially, a temporary set of "sky objects" (purportedly areas of blank sky) are added to the coadd and their PSF fluxes are measured. This is used to scale the the nominal detection threshold to an "effective" one that better matches the S/N requested (using the ratio of the empirical RMS of the sky object flux to the mean of its quoted uncertainty and setting hreshold: effective = nominal*err/rms). Further, in DM-13741, an additional option (defaults to True) to tweak the background such that the distribution of the sky object PSF fluxes are centered on zero.
Recently, the HSC folk have reported that the persisted sky objects (which are different from the set used in dynamic detection, being added after the detection phase solely for post-processing QA) have a distribution that is centered at a non-zero (in the positive direction) value. This should not be possible if the above background tweaking is working as expected. This requires investigation.
Given that the sky objects used in the dynamic detection/background tweaking phase are not persisted, this will involve either including some --debug functionality, for in-situ analysis, and/or persisting those sky objects for study post-processing.