There is no one size fits all.
I like “auto select star” and “ use subframes”
Here is my setup and routine, working with a diy arduino based, 3D printed focuser.
Inherently this little machine has a fair amount of backlash, which is often the reason for failed processes.
Anyway, the linear process, implemented fairly recently deals with the backlash issue the best, IMHO. Starting from a “last good” focus position it drives to *in my case* inside by 300 clicks, measures FMHW every 30 clicks while driving through optimal focus to outside....and establishes a v curve. After return to inside it “walks back” to optimum focus, in about 80% of the time it will complete successfully after 15 steps.
Variables I had to adjust: different filters will require different exposure. For a 125 F6 refractor, with an ASI1600MM I typically use 2 sec for L, and 5 for RGB. Narrowband I found to be tricky, sometimes depending on the target I may get away with just 5, but often I have to bump this up to 10 or even 15 sec,
The value for the initial move from ”last good” to start the establishment of the V-curve is 300 clicks, roughly 10 times the critical focus zone. Narrower than this resulted in failed focus procedures due to inability to establish a good v curve, wider than 300 resulted in inaccuracy...
I spent several evenings playing with these values, and making sure I get a nice V-curve with any filter.
I second Jabians point about starting with a non automatic focus first. I stop the autofocus routine once the first image comes in, select a star myself, and then hit “loop” which now just brings subframes that look pretty identical on the screen. Using the arrow buttons I drive the focus in or out to get the smallest possible star image. There i stop, maybe a für a few back and forths, and initiate autofocus,..
Works for me quite well... attached a current screenshot, hope this helps...