Last night there was some nice weather and I decided to run a bunch of tests of a current driver. Most of my previous comments and conclusions stands, but I have a little bin more data and some additional conclusions and comparisons.
It was run with astroberry compiled caux and other drivers and kstars compiled from git.

• My setup was deliberately sloppy - no precise levelling or pointing. I just moved my mount to the balcony, connected, it levelled the tube with marks on the mount and started.
• First, I checked the "point to the moon" scenario - it does not work. Here we are worse than skyportal/skysafari. It <em>should</em> work if we would have all our ducks in the row in the alignment subsystem. It should be able to handle the rigid rotation in both axis (shifted zeros) without any problems. Something is wrong either in my handling of the plugins or in the plugins themselves. So Maciek was <em>right</em>. Jasem, please during your refactor pay attention to my usage of the alignment routines. They are programmed after one of the tutorials, but maybe I did something wrong. BTW, instructions of skyportal tell us to do the same and I remember it got confused few times when I didn't.
• The alternative scenario: point to the northern horizon (my pointing was +/-10deg approx), do a goto some object, center it and sync, repeat two more times. This works fine even with visual centering (no plate solver). The result was adequate for visual observing.
• A similar procedure with plate solver and sync points (30-60deg) around the target worked much better.
• By doing 3-7 sync points, I was able to reach the field rotation barrier with my expositions (60-120s).
• I run the setup alternatively between skysafari and indi by disconnecting the  CAUX driver and connecting SS. Without dropping calibration in both. It worked for at least four cycles. I synced the positions in SS on bright stars and live view in the camera. The first was a bit tricky, but the rest were easy. I had to go outside only once during the run.
• There seems to be no large qualitative difference between these drivers. The SS is a bit smoother in operation (being on the tablet). There seems to be some quantitative advantage to SS with alignment. It is more tolerant and the accuracy falls off slower with the distance.
• Both internal and SVD plugin worked similarly with a small edge towards internal. The nearest plugin did not track.