The defining feature of my pointing calculations - and their inconsistency - is that the spread for individual sources is bad, which means that there's still something wrong in the way the bulk ra/dec are being calculated.

I checked apply_distortion_map_radec to see if it might have inserted some bulk offset, but it only changes the average position by <2". That might be an issue, but the boresight isn't necessarily aligned with the mean location of the bolometers. It should be shifted by at least .3" according to a simple average of sin/cos of the bolo_params angles. Comparing my pointing code to map_ncdf_reading directly:

readstruct=map_ncdf_reading(filename,/nopixoff) offsets are small (e.g. -0.050665803 -0.0029540043 arcseconds ra/dec) if I don't subtract fazo/fzao offsets in ra/dec are remarkably close to fazo/fzao if I subtract them:

92.830656 -126.12773
95.000000 -120.00000

which is because ra/dec are pretty closely aligned to az/el. So, any difference comes from at least one of the keyword parameters. Therefore I AM missing something, and that something is one of the keywords to map_ncdf_reading