Take a look at the DONUTS article I posted above, ATrack, a software program I developed, follows closely what they did. In other words, to maintain the very high tracking accuracy, ATrack does both drift correction AND re-centering. After three or four images ATrack has corrected the drive rates to achieve tracking errors less than 1 arc-second per 60 second images. But small errors in drift can add up and move the image away from center. But re-centering done very carefully after each science frame (see DONUTS article) maintains an absolute centering over several hours, on my system, of about 0.2 arcsec for both RA and DEC.
My location is in New Zealand, and the seeing is about 2.5 arcseconds. I use Pinpoint to provide very accurate plate solves which gives subpixel accuracy (after I fine tuned the plate solve parameters). The OTA is a GSO 16" RC, and the optical train behind the mirror is a Moonlite NC and a SBIG 6303 camera and FW. If this system was poorly designed then I would see some significant random jumps in the plate solves .. I do not. So the system is very solid and predictable .. which is why ATrack works so well.
For example, last night I did four photometry runs: HA 0 to HA +3, HA -3 to 0, HA 0 to +3, HA -3 to 0. In all four cases the ATrack drift rate corrections for RA and DEC were pretty much the same, so it didn't matter the DEC or the HA. And in all four cases the difference from the last frame and the first frame was less than 0.5 arcsec. This shows the stability of the entire system, including the mount, OTA and instrument optical train at the back of the mirror. Temperatures went from +10C down to +3C by the end of the night. And there was a full moon up which really didn't affect the imaging much due to the very small FOV and angular distance from the moon. There was no significant cloud or fog in the sky to scatter the moon light. So overall, a very successful night.