The templates in the ap_verify_hits2015 dataset are psfMatched coadds from the HiTS 2014 fields. The images selected to be coadded were chosen using good seeing criteria. Until recently, these were all the AP team had been using for image differencing templates. Now with
DM-15370 and DM-15371 done, there exist direct, compareWarp, and DCR coadds with the same constituent images. Further, ap_pipe has been run on the HiTS 2015 dataset using each of these as templates.
When DIAObjects from the PPDB are plotted on the sky, the rerun which used the psfMatched templates exhibits a significant amount of large-scale ring-like structure across the decam focal plane. This structure is absent in all three of the others. In addition, the psfMatched coadds appear to be a lot messier (by eye) overall than their non-psfMatched counterparts.
This ticket is to figure out what is going wrong with the psfMatched coadds, because in principle, PSF matching should be a good thing. A notebook showing the gist of the problem is available on GitHub.