Good info. I will add this:
1) The docs recommend various rejection schemes based on number of subs.
2) To improve SNR the idea here is to reject as few pixels as possible, only those involved in noise structures. Jordi Gallego as a good tutorial "Image integration techniques" on this
page. For my integrations I usually leave sigma low at 4 and increase sigma high from 2 to 2.4 or so.
3) Good question. I have been using the sub with the best subjective compromise between the least FWHM, the roundest stars and the least noise as measured by the NoiseEvaluation script.
Mike