While trying to track down some bugs on DM-641, I've grown frustrated with the difficulty of testing the deeply-buried (i.e. interfaces I want to test are private) shapelet evaluation code there. That sort of code really belongs in the shapelet package (not meas_multifit) anyway, where I have a lot of similar code, so on this issue I'm going to move it there and refactor the existing code so it all fits together better.