Thanks for thinking about this off-the-beaten-path project.
Choosing "offset init" rather than "cal star init" has no benefit, even when calibration points have been cleared. Also there is no benefit setting PID to (15000,1000,12000). That is, oscillation triggers when the scope is moved after initiation.
The oscillation remains remarkably consistent: "error" in the numbers tab climbing to ~12' in azimuth and snapping back to 0. Plus there is that odd virtual handpad behavior in azimuth: the scope only rotates clockwise when SiTechExe is started -- that is, even when the left buttons are pressed, the scope only rotates right -- and then if the oscillation is triggered, the scope only rotates counterclockwise, even if the right button is pressed. Finally, there was that strange "awakening" where, for a short time, the virtual handpad worked properly and oscillation was not triggered.
I don't need great tracking. Say, less than 2 arcminutes error in 2 or 3 minutes for right now. I do need precision pointing, bringing faint objects to the center of field of view -- I had planned to achieve that with successive approximations using plate solving. It seemed that simple offset init didn't quite get me the tracking I needed but that 3 cal stars did. Perhaps it would be enough to repeat the offset init when the scope is close to its target?