thanks for sharing those images with me - being able to dig into those let me work out what is going on (which is image specific to a certain extent). There were two issue
1) SharpCap was performing star detection and finding the 1000 brightest stars in the image before filtering those stars to select the ones that are suitable for FWHM measurement (ie not saturated). In the case of those images, the 1000 brightest stars are in some cases all saturated (or there are a very small number of non saturated stars). This could lead to no FWHM reading in some cases and to a spurious reduction in star detection sensitivity due to a small logic error. Changing to detect 2000 stars before filtering gives a good selection of non-saturated stars, which increases measurement stability and reliability
2) The filtering of which stars were OK to use for FWHM calculations was slightly incorrect in live stacking - it was checking for saturation of any pixel, but on a slightly smoothed version of the star brightness data. This makes a big difference on the small FWHM stars in the images since even a slight smoothing can pull down the peak level significantly. With somewhat broader stars in the 3-4 pixel FWHM, the effect is far less noticeable.
As usual all of this will be resolved in the next update
cheers,
Robin