In recent processing runs, detectors have been failing with issues relating to the PSF evaluation at certain locations. There is a config to "catch" these errors and allow the SFP to continue. The question is then whether it is safe to turn on this config (i.e. if the calexp is truly a hopeless case, can/would it get flagged and not included in the coadd based on our input selection criteria). Investigations on DM-36843 revealed that these detectors actually do pass our current checks. This and the analysis on DM-36930 have highlighted the need to add additional metrics to be thresholded for consideration in coaddition (and potentially some robustification of PIFF itself).

One situation where the PSF modeling can go awry is if there is not good coverage of PSF model stars over the entire unmasked region of the detector (due to the resulting need for extrapolation of the PSF model, which seem to be quite unstable, see also this comment on DM-36930). This ticket is to add a metric that computes the maximum distance from an unmasked pixel to a PSF model star in a given detector. Its effectiveness as a metric for identifying problematic detectors that should not be included in the coadds will start with a close look at all the failed detectors noted and fully described on DM-36930.

A code snippet for the computation of this metric was provided in DM-36930. This will be adapted and added into the ComputeExposureSummaryStatsTask in pipe_tasks (noting that it may get moved around pending the implementation of DM-35207).