These nice before and after plots shown in
DM-13410 were from a the pre-review test run. The version that got merged doesn't set SENSOR_EDGE (and consequently INEXACT_PSF) of the new EDGE pixels.
This has the effect that fewer sources measured on the coadd will have the inexactPsf_flag set. However, most (80-98%) of the EDGE pixels are also BAD or INTRP which are then "REJECTED" which then propagates to INEXACT_PSF: