Hi Paul and Fukinagashi,
min exposure absolute value of camera is determined by driver. You cannot go under. You can define the interval of min max exposure for flat in flat configuration but you cannot go under the absolute min value.
Generally better to not use too small ROI for flat calculation if you have big sensor and/or really not linear light decrease at corner (for example fast F ratio telescope). For CMOS camera you can use also the full frame if the sensor isn’t a monster sensor.
About the tollerance for my skyflat I use 18% of tollerance. Really important thing in this case is to normalize flat before doing rejection and median (I use CCDStack for this and its really simple).
This is My SkyFlat at Dawn, small number of flat for all filters … in few night I have a huge library of flat. I redo the master flat each 3 months with the last 3 months flats file. CMOS of actual generation do not need Dark Flat so I use the Bias for calibration (some older like the ASI 1600 sensor need dark flat).
For SkyFlat … absolutely do not use flats taken under heavy veils or cloudy sky.
All the best
Leonardo