My last session with this was mostly fiddling with the XY table without any clear victory.
But first check out some curved surface finishing with fine step-over. 10% vs the 40% already shown in prior log:
Resetting the milling bit more carefully reduced runout which did clean up the pocket walls that I groused about last time, where the round thing closely approaches the middle of each side wall. In the photo above, the pocket wall to the upper right part of the left/latest pocket looks pretty flat and clean (wrt my expectations). The internal reflection of the slotted corner beyond it - the slot just visible in the crop above - shows clearly through the new pocket wall. More clearly so in person than the photo shows. I'll call that acceptable for this project. The wall opposite to that, which you can't see except for the top edge, looks pretty good too.
But the craggy wall in the upper left corner of the picture gives evidence of unsolved problems. Also the big square facing cut was less clean than the prior iteration, and both were less clean across the "top" (picture orientation) quadrant. This time I spotted that the X axis was chattering while moving one direction (as in the "top" quadrant) during the facing operation. Ok, so maybe I got greedy about squeezing out backlash. But both the craggy and clean visible pocket walls were cut with the X axis moving in that direction, so there is some cross-interaction too.
By the way, I thought some of the irregularity of prior tests might have been due to an axial pre-load spring in the motors yielding to force transients. So part of the last session was adding screws to pre-squeeze that preload, using pilot holes that I'd CADed into the laser cut parts but hadn't tried to use yet.
I did a lot of fiddling with trying to get that and the anti-backlash squeeze between the lead screw follower and the driven part of each axis tight but not binding.
The "chatter" problem appears to happen when an axis is yielding to force applied in the direction of motion. In other words, permitting the axis to move at a measured rate rather than pushing against resistance. This first turned up early on as the reason for arranging this machine vertically with a counter-weight instead of moving the heavy spindle horizontally like #CDCNC . When I first turned the XY table sideways, with Y vertical, it easily drove the Y axis up, against gravity, but chattered badly or stuck (lost sync with fast stepping) while bringing the axis down, with gravity. I tried a few things at that time including light oil and different nuts on the lead screw. Different combinations gave different results, varying in degrees of bad. So to keep moving I put the XY table horizontal and rigged the counterweight to offset the weight of the Dremel-clone spindle in vertical motion.
After reading a bit, last session started with trying PTFE dry lube to see if that would fix the chatter-on-descending and allow orienting Y vertically. Mainly since discovering appreciation for how the horizontal arrangement allows cut chips to fall off the workpiece. It did not.
At other points of adjusting the X & Y axes, one would move well against a small resistance but chatter and/or bind otherwise.
So at this point, it would be great if some better understanding of constrained-nut-on-spinning-screw dynamics happened to fall from the sky.
Some periodicity appears in the face cut (the left big square) shown above. Very clearly along the "bottom" (picture orientation) edge. Cutting that area seemed to go smoothly, so that's not the disruptive chatter problem. The period measures 1 mm. That's two turns of the lead screw. Unless nearness to 1 mm is simply an unlikely accident, I can't think of a source other than the lead screw. And I don't know how the lead screw would generate an effect every two revolutions. ??.
Bonus video for whoever reads (or skips) to the end. More useful for self-hypnosis by CNC rather than for new information.