DESC reported that running jointcal on Run2.1i did not improve the photometric repeatability (esp. wrt the reference catalog) relative to processCcd. This is not surprising because:
1) The reference catalogs are near perfect and to full depth < 23. They essentially are the input catalogs, and therefore jointcal's benefit of bootstrapping deeper stars is not realized.
2) There is no vignetting or sub-ccd throughput variability in the imsim images in Run2.1i. Therefore, the photoCalTask model of "1 zeropoint per ccd" is the correct model, and jointcal's ability to have variable throughput across the focal plane and per ccd is not needed.
However, various test plots (either via James Chiang or the visitAnalysis scripts) show that the calibrated photometry is only consistent with the reference catalogs at the 13-16 mmag level, which is much larger than I would expect for clean simulated data for bright stars (basically, I would expect few mmag consistency).
This ticket is to describe work to see where in processCcd/jointcal things are going wrong.