Lilo I am not yet convinced that it is a glitchy level sensor, although that is a top suspect. It may be that backlash is contributing to this but I need to do some more testing.
Maybe ZWO Tech Support can let us know if when doing the 1st set of 3 plate solutions the Seestar takes out the backlash when it goes to the 1st point. Since the RA axis travels in the same direction for the 2nd and 3rd points backlash should not be a problem.
The next question would be, if we back out of the PA routine and came back in as marvindj mentions above, does the Seestar take out the backlash when it moves in the opposite direction in RA to get back to the PA starting position. The difference between the first time PA and the second, third, etc. is where the RA axis starts so, if backlash has not been taken into account this could be one of the sources of variability when we repeat the PA process. And keep in mind, that not every Seestar will have the same amount of backlash as this will be a variable in the manufacturing process. So some folks will get more of a repeatability problem compared to others.