I think everyone is having trouble understanding me because what I'm saying is so obvious it is hard to explain. Forget the pointing model. The mount is asked to go to x, y - and it lands at x+dx, y+dy - according to its own model and its own encoders. It knows it made a mistake in the end - but it's hard to do it accurately.
Imagine a missile with extremely accurate - perfect - gps. You fire it at a target location and it misses by 10 meters to the right and 10 meters down. In its last moments it knew it was in the wrong place - but it was too late.
If you are saying the mount is bad or dumb - keep in mind it is not intended for 10" accuracy in a goto. It is intended to point accurately across the sky - and since it is ultimately limited by the mount model to 3-5', some amount of landing error is in the tolerance.
Keep in mind that from the mount's perspective, software that can't adjust for this tiny little matter - either by nudging or the few lines of code I suggest - is the thing that is broken. The mount is performing to its spec very well. The software is designed to center targets very accurately using plate solves - and indeed it could - but it doesn't because it makes a particular demand of the mount that is unnecessary. And it insists on syncing, which is also problematic and unnecessary.
Frank