The graph shows clearly that a small change in servo sidereal rate will result in a significant drift in RA, 10 arcsec over 700 seconds.
Last night ATrack showed an average drift in RA of -0.01500 arcsec/sec. The drift range for RA on my system is always between -0.01200 and -0.02000 no matter what the temperature or where the scope is pointed. It says the RA tracking rate is always too fast. The range is most likely due to refraction, polar mis-alignment and other mechanical deviations. But the underlying base (e.g., -0.01000) is constant. Using an average of -0.01500 arcsec/sec means a drift rate of 10.5 arcsec over 700 seconds .. very similar to what you see in the graph above.
Even more important is when I use the APCC model correction it has a different value for the RA correction, BUT it tracks in parallel with ATrack correction. Meaning, the two corrections are in parallel. This would indicate a tracking error constant that APCC is not accounting for. The likely reason is a 0.1% error in the servo sidereal rate. But I would also agree it could be something else. If the model is not accounting for actual mechanical changes in the mount then there will be a divergence over time between the APCC model tracking correction and ATrack tracking correction.
One thing I can try, ATrack has the ability to provide a "base" correction to the motors, this is to take into account a constant between the motors and the real time tracking. On my next clear night I will set the base tracking rate and then turn APCC tracking correction back on. In effect this will simulate changing the servo sidereal rate which is then modified with the model correction rate. I should see a much improved tracking anywhere in the sky, depending of course on how accurate the model reflects actual sky conditions. But it should be better than it is without the base correction.