NSG, after 3 hours, tells me that only one of my 463 subs doesn't suck

Jared Willson:
Likely, you had a bad reference frame in NSG. I use WBPP to cull bad frames (by setting a minimum eccentricity and FWHM), then I sort my frames by SNR and figure write down a couple frames with high values (but not suspiciously high) and also look at the altitude. Generally, the SNR should improve with altitude as long as there is little or no cloud cover. As long as there is general agreement between SNR and altitude, I'll pick one near the top--doesn't matter too much which one--and write that one down. I would NOT recommend skipping normalize scale gradient. Even though it is a bit slow, it does wonders in terms of simplifying gradients for DBE, and my testing suggests I get consistently higher (if slightly so) SNR in my images when I use it with almost no cost (typically less than 0.1 arc seconds) in resolution. 

My guess is you chose a reference frame that, for whatever reason, measured much higher in terms of SNR. As a result, nearly every other frame received a weighting of less than 25%, so they were all dropped out of the integration stack. That's an indication of a problem, so I would not recommend just integrating anyway, at least not with the NWEIGHT FITS keyword as the weighting source. You could use your SSWEIGHT instead as the weighting source and still integrate all your normalized frames, or you could start the normalization process again with more careful selection of a reference frame. Perhaps choose whichever frame was second best? Then double check to make sure you don't have any frames (your prior reference, in particular), that suddenly has a weighting of 8 or 10 or something that would throw off your overall results. You can see the weightings in the pre-pend for the names of the normalized files.