I cannot claim credit for inventing it - I think it was thought of before but not well documented. It just struck me as an idea and the PI gurus confirmed it made sense and had been tried before.
Simply put, if we do LRGB imaging and use LRGBCombination, we throw away the luminance information in the RGB image. It just occurred to combine all four stacks to form a superlum. Since all the stacks already had their outliers rejected, I just weighted their combination according to their noise level with ImageIntegration.
My noise levels dropped dramatically when I did it on M3 since I had an equal number of 5-minute subs for each filter. If you use RGB sparingly, as many do, the benefits will be less significant.
The documentation will be included in Edition 2 of my book. I'm hoping to get enough image acquisition to support a publish date of 2018