Just a few comments Kevin,
1. A paper disc with a single fine reference line attached to the motor shaft can provide for comparing
actual steps to commanded steps. Poor mans mechanical encoder for checking if steps are lost ( your trying to isolate the commanded steps to physical movements delivered ).
2. No backlash implemented in your xml. The closer you try to get to perfect the harder it is to find the
what is causing the out of tolerance. AT .05mm ( .002" ) your in the area that a number of things can be
adding up to the tolerance call it the "system influence" . Lead screw tolerance, backlash, your measurements, etc.
Just thinking out of the box FWIW,
RICH