I decided to try one for myself as i was sceptical that it was all that bad - also was very excited about the possibility of using the new Sigma DN Arts and Contemporaries.
I tested it properly in October this year in Crete on the dense starfields of Cygnus and was absolutely shocked at how bad the data was.
Modded Sony A7II + Sigma 65mm Contemporary @f/2.0 - 100% zoom center crop from a debayered raw :
It has completely mangled the data. Ive chucked it in a box at the back of the astro spares cupboard as i would feel guilty selling it to someone as an astro camera its so bad
I own 4 modded Canon DSLRs and the data from them is immaculate in comparison.
I used to use a Sony A7s, mainly for timelapse nightscapes, but out of necessity I used it for regular tracked nightscape panos a couple times. I paired it with a Voigtlander 50mm APO and a Voigtlander 21mm Nokton, I got pretty much the same results as you did. Downsampled and wide? Sure, looks like a good picture, but once you look at the individual frames at even a modest zoom level the star mangling is obvious and disgusting. This was the best I could manage from several panos that got destroyed by the star eater, when people deny what the algorithm does because "I don't see what the problem is, I get good photos" or "there's already too many stars anyways, what's a few missing going to do?", I just laugh.
Here's a link to a single, unedited, frame that is part of that pano, zoomed out it doesn't look terrible, at 100% it looks absolutely aweful: