I've used my simulated data and confirmed that there is in fact an error in the l/b mapping, but I haven't tracked it down yet. Stay tuned. Update: Something is not self-consistent in prepare_map. I've tried a variety of different things without success yet.