May be the same issue I reported here a few days ago..
https://bbs.zwoastro.com/d/17780-tracking-drift-starts-only-after-start-capture/12
I think there is about 1/4 frame width worth of gear lash (total slop in the gears) and whether the last movement it makes to find the object after identifying position comes from the left or right affects how far it drifts as it takes up that lash.
In the northern hemisphere, facing south toward the ecliptic, the whole sky is always moving left to right at 1/4 degree per minute. What I find is that when I make manual movements of the scope panning from right to left (causing the astro object to move left to right in frame) and then stop, it'll start drifting for a while. It thinks it's tracking fine, but it's just driving through the gap in the gear lash. During that period the gears are turning but the scope is not, so the sky starts drifting across the sky from left to right. Eventually it'll hit the end of the lash, and then track normally, but the object may not be centered any more.
My suspicion is that whether it starts drifting right away or not depends on which direction is was moving after it does its three calibration points (on first capture) and returns to and identifies the object. If you watch in the Sky Atlas it sometimes shifts the box around a bit as it's identifying. I think that if it makes its final adjustment moving the scope right to left (opposite the normal tracking direction), then it will drift. If final movement is left to right (same direction it tracks) then no drift.
My workaround for this when I see it, is to manually drive it to to the right a couple taps in slow mode to take up the lash, and then it seems to track normally after that. Since my movement is the same direction that it will be tracking, there is no additional lash. My manual workaround could be implemented in software.
That's my hypothesis and the workaround has helped me get a quicker initial capture. Beyond that, we know it does an intentional movement for purposes of dithering every minute or so, and I think it actually reidentifies/recenters the target area which may be why it seems to just "fix" itself after a while of drifting. Also after I've let it do the 3 point horizontal calibration, I usually just disable that feature, and it's quicker to move between objects without additional bouts of drifting.