Jim and Andrew, those are very very good ideas! I think both of those could make sense. We should definitely take a look.
I just made a change to the autodownsample function for the solver, to just be a little less aggressive because it was causing some failed solves since the stars were downsampled too much and were no longer starlike.