Hi Jack,
The fundamental premise behind all 'K-Sigma' sliders is that the software looks at the image data and establishes the 'Mean' (or Median) and the Standard Deviation (StDev).
The 'StDev' represents the amount of 'spread' of the data, and I can't remember the exact maths here as I am not near my reference books, but a 'single' StDev (or Sigma = 1) should contain (and this is purely an 'example figure' here), say, 95% of the data.
If you apply a 'multiplier' (Sigma = 2.00) to the StDev figure, then you are telling the software that you want to include MORE of the OUTLYING data - so a Sigma=2.00 applied to the above thought experiment might yield 96.3% of included data.
However, if you have some data that IS 'outlying' at the very first inspection, then this data will have 'skewed' the original Mean or Median, around which the (Sigma x StDev) capture window will be centred. This is where Winsorized Sigma Clipping comes in to play - as it can start to 'intelligently' exclude these outlying data values, providing a more and more 'robust' estimate for the true 'Mean' of the data set as the algorithm 'loops' (or iterates).
And many of the PI algorithms that include 'Sigma' also allow Sigma to be independently specified for the 'lower' and 'higher' ends of the distribution curve - very important as our astronomical data tends NOT to be equally distributed on each side of a mean value.
All that said, I am basing THIS response on my more detailed understanding of the ImageIntegration process, NOT the ATWT process that you initially asked about. However, if you just bear in mind the fundamentals, it may help:
1.) Analyse the data set
2.) From the statistical analysis, determine the Mean and the StandardDeviation of the data
3.) Expand the 1-Sigma deviation (StDev) by a factor, K, such that the included data is represented by (K x StDev) - often just abbreviated to 'Sigma'
Cheers,
