So I have some more information. Not sure that it helps with anything, but more information. I had ran a backlash routine before that dc42 had posted. But it was running in more in the center of the bed. I've been stuck thinking backlash was only in 'axis travel' and not somewhere else. That is poorly worded.
I decided to take a different approach after someone said it looked like one axis was causing it. I had assumed all 3 were, or I guess it could just be the rear tower as you don't see the problem there as much because of the probing pattern. So I wrote a program to probe at a radius of 100mm in line with each tower 4 times after an 30mm outward move then 4 after an 30mm inward move both inline with the tower. Here are my results.
12/2/2019, 1:42:41 AM: : G32 bed probe heights: -0.023 -0.017 -0.017 -0.017, mean -0.019, deviation from mean 0.003
12/2/2019, 1:43:13 AM: : G32 bed probe heights: -0.130 -0.136 -0.136 -0.136, mean -0.135, deviation from mean 0.003
12/2/2019, 1:43:46 AM: : G32 bed probe heights: 0.045 0.045 0.045 0.045, mean 0.045, deviation from mean 0.000
12/2/2019, 1:44:19 AM: : G32 bed probe heights: -0.111 -0.111 -0.111 -0.111, mean -0.111, deviation from mean 0.000
12/2/2019, 1:44:51 AM: : G32 bed probe heights: 0.040 0.040 0.040 0.040, mean 0.040, deviation from mean 0.000
12/2/2019, 1:45:24 AM: : G32 bed probe heights: -0.110 -0.117 -0.110 -0.110, mean -0.112, deviation from mean 0.003
So there it is. Now if I move up and down with an indicator at one point, I don't see even a quarter of that.
Looking over the machine again, I noticed movement in a skate. I went to snug a screw and the bearing journal post inside broke off. Found cracks in all of them. I had replaced them earlier this year and emailed seemecnc if I could just buy the shells. They sent me replacements at no charge! Great people!
So that could have been the problem, or not. I probably over tightened the screws. Going to replace them with 3/4" long instead of the 1/2" provided to keep from putting all the stress at the weakest point.
Hopefully that was the whole problem all along. I would think any other backlash would show up in my previous tests.