The Sgr B2 ALMA data from project 2013.1.00269.S has imaging problems around the brightest sources, especially Sgr B2 M. Crystal Brogan & Todd Hunter gave some useful advice, essentially the following:

• the line forest is causing trouble, so line-containing regions of the spectrum must be excluded when mapping
• there is some nasty curvature in the spectral baseline that needs to be removed probably by flagging out some of the data
• Self-calibration is needed to exceed a dynamic range ~100

Todd developed a fitcontinuum.py script to address the first point, but I ended up doing this custom (script) because of the problem introduced by the second point.

To exclude the lines, I had to use frequency-based selection, not channel based, because my spectral windows from different dates were offset by 10 MHz. The approach to use in CASA is ugly: you flag the line-containing data, then split (and in splitting, average) the continuum into another MS. From there, a phase-only selfcal is straigtforward (script).