Currently GetCoaddAsTemplateTask sets the PSF of the template with that of the first warp that it adds in. Bob Armstrong improved this on https://github.com/LSSTDESC/dia_pipe/blob/master/python/lsst/dia/pipe/getMultiTractTemplate.py. However, he reports an edge case where it fails trying to compute the average position of the composite PSF near edges because the average position can be in a region of no data.
Bob Armstrong will ask Bruno for tract/patch that failed as a how-to-reproduce.
This ticket covers ensuring that the PSF is valid across the whole template