Can you give it a try with this version:
I think that this should (hopefully) fix your issue. The problem was that the same current threshold was being used when retracting the belts and when taking a measurement so changing the retract belts threshold was also messing with calibration.