I think just to feel better about myself, I'll stack the SN hour in with the prior data. That way I'm just replacing things (supernovae, spaceships running away full throttle) removed by sigma rejection.
I'll slowly get my head around NewSVD with time, feeling out how it reacts to different datasets and processing. I do most of my adjustments within SVD to the best I can manage right now, so any subsequent artifact repair is minimized.
One of the first things I did with PI was to turn off cosmetic correction for stacking, but yes it is likely the bigger stars have some full white pixels in there. My L subs are only 30s, but even on 10s tests I can saturate a few pixels of mere run-of-the-mill stars. Why these cameras don't have an SD card slot for expanding the full well I just don't know.
Most of my data here, despite being over two (soon three) nights, came in with very similar statistics, so they shouldn't be "too" funky. But I appreciate the explanation of what SVD might be turning up in data mining my stars. Makes me wonder if I have any program that could give me a graphic representation of a chosen star's PSF (like a 2D slice).
Indeed for the weak 1-hour SN stack, I did back off sampled iterations to 5, which helped as to both ringing and noise amplification, but still otherwise was improved over pre-SVD.
Interestingly, the sample screen offered me the SN as a clickable starfishy. Query - is it really a point source anymore, or just little pieces strewn out all over? Billions and billions of pieces.
It's an area that's definitely worth revisiting if you ever end up going higher magnification and narrower fields.