The algorithm that selects guide stars isn't perfect, for sure. It tries to find stars that are reasonably bright, but not totally blown out. It tries to find stars not really close to other stars. It tries to find stars not right next to the image border. Here it is:
invent.kde.org/education/kstars/-/blob/m.../guidestars.cpp#L139
If you lowered the StellarSolver min size parameter to 0, I'd guess the scheme would probably start detecting some bad pixels or noise as stars, and perhaps sometimes those stars would be added to the mix and could affect guiding. Real stars that are that small are also probably not very good guide stars, as they probably are harder to detect consistently. However, if all it detected were small stars, then for sure it should be using them, instead of sticking to an arbitrary cutoff of MinSize=1.5HFR.
Like everything else, it can be better and I occasionally get back to it. Probably the key thing needed to improve it is some idea of what stars look like
for your particular camera and scope and exposure time. I suppose I could get it to take several "practice exposures", evaluate detected stars with a variety of settings, find ones that seem best, and then adjust settings accordingly. Some day I may get to that
For now, it's best if you adjust your StellarSolver settings so that it does a good job of detecting stars for your setup. The hope was that the default StellarSolver settings would do a decent job, and it probably has, but if you see signs of issues (like these dropped guide stars) then if you know how, make adjustments to the star detection parameters. Obviously, be conservative with your changes, try your best to understand what the parameters are, and write down what your old parameters were so you can revert back to them.
Hy