Jointcal is currently highly sensitive to the minimum S/N cut imposed on input catalogs. Slight changes in S/N cause the relative or absolute errors to go out of bounds (on the lsstSim tests in testJointcal.py). Either those errors are too stringent, or we need to figure out how to make jointcal less dependent on the details of how sources are selected for inclusion.

It may not be the S/N cut itself that causes the problem, but rather other properties (e.g. covariance of centroid) that are related to the S/N cut. But the S/N is the easiest property to tweak.