I ran the v0.7 reductions with deconvolution on for 50 iterations. I had cut out deconvolution originally because of the funky noise maps, but that was partly an error on my part. There is also an issue with bright sources being largely left over in the noise maps. The deconvolver does MUCH better at filtering out the fuzzy atmospheric emission, so I want to use it. It leaves some flux from bright point sources behind, though, so I decided to try to make the deconvolution kernel smaller to see if that recovers more of the pointlike flux.