if you see 0.25 then they are hot pixels.
the reason being that most DSLRs have 14-bit a/d converters. when PI opens one of these files, it puts the 14-bit data into the lower 14 bits of 16 bit integers. therefore the max of 11 1111 1111 1111 is represented as 0011 1111 1111 1111, which is 16383. 16383/65536 is about 0.25.
as StarAlignment is running, it prints a transformation matrix for each sub exposure. after that matrix it also shows the "human readable" X and Y offsets. this stuff all flies by in the console so you'd have to scroll back to find it.
if rejection was not turned on, then there will not be a high rejection map. if you have enough subs (~15+) the best option is to use Windsorized Sigma clipping. i usually end up with both low and high sigma sliders in the vicinity of 3, but you need to inspect the rejection maps to see if there is any structure of your object in them... if so, the rejection slider(s) are set too low.
usually that kind of steady march in X/Y in your subs is the result of differential flexure between the guider and imager. if unguided, it can be the result of inexact polar alignment. in either case, sometimes nearby hot pixels start to overlap due to that diagonal motion and then they become hard to reject. this is the source of the "rain noise" that's often seen in DSLR images that were undithered.
if you end up there, then cosmetic correction is the way out. note also that sometimes due to dark scaling during calibration of your images, that hot pixels are only partially corrected, and end up *becoming* warm pixels... which due to the fact that they are not strong outliers anymore, can also be hard to reject.
rob
