For what I can see in your curves, you should have an apreciable wobble. int't it? My feeling at this point is that the multiple errors are adding in a complex way, so a single rotation of the worm is not enough to get all the curve profile. I've the same problem with my mount.

I you look at the curve before you start recording, you can see that two consecutive periods are not equal, so recording corrections for one of them will make works the tracking on the second one. Saddly I think think that the PEC will not be completely effective until we remove all the eccentricities from our pulls and belt trains. In some periods it will cancel a good amount of deviations, but in other it will be ineffective or worse :(