Bob, I have no idea why you would not believe me about what backlash I'm seeing. I'm telling you what I see on multiple machines which is also what Taig says it will so. It sounds like your just pushing things beyond their mechanical design.
Obviously because I am a total idiot and incapable of doing simple experiments like cranking down everything until it stalls and then backing it off enough to run even to do a simple no cutting load test of backlash under ideal circumstances.
I did exactly that yesterday. I started by turning backlash comp off obviously. No load. Cranked down nuts until it stalled, then backed them off. Did the same thing with gibbs. Mobil way oil on the saddles and spindle oil on the screws to give it the best chance of good results. Found there was a some backlash in the bearings so I replaced them with exactly the same part number bearing (I had spares on hand) and found there was still some back lash in the bearings. Sorry, cheap skate bearings are going to vary, but I knew that. Then I did some rapid change tests and found with everything tightened just so and well lubricated there was still about .0005 to .0007 just in the plastic pin couplers. Less if I loosened up the table, but then what's the point. LOL.
I also did some cycle tests to make sure of repeatability. The machine was not losing steps. With a hundred cycles on each axis repeatability was sub .001" while I was testing. Tests were run travel limit to travel limit on X & Y.
I took your word that is "possible", but I still doubted the hundreds of hours claim. With relatively new (less than a couple hundred hours) screws and nuts and new bearings it checked at .0015 on X and .0025 on Y apx with everything as tight as I can make it without losing steps. Checked with two different dial indicators. One was a Starret. One was a Shars import.
Now maybe I have a lemon, or maybe I am really am an idiot, but after spending all afternoon experimenting (not the first time) I do have some difficulty with the the sub .001 for hundreds of hours without adjustment claim even with light cutting loads. I'ld be hard pressed to believe it is possible on this machine atleast even with virtually no cutting loads. I can probably get it under .001 with ACBs and helical flex couplers, but I would still be doubtful about being able to keep it there for hundreds of hours without adjusting it. I could be wrong though. I hope so. The worst of the backlash does appear to be in the bearings. Since I couldn't find ACBs that small with grease seals I'll have no choice but to set up some kind of oil mechanism on them.
I would note that the Z axis way plate on the machine was never straight from the factory, and I had to shim it in order to get it close. Maybe I did get a Monday morning Taig, but the the design just doesn't impress me that its capable of hundreds of hours of actual cutting with sub .001 backlash with stock components without having to readjust it.
When I got your "it must be you" comment I decided to test it to the best of my ability. You have setup a bunch of these afterall, and I have setup one.
BUT NOBODY WHO HAS RESPONDED TO THIS HAS YET TO GIVE A GOOD ANSWER ABOUT WHY THE BACKLASH COMPENSATION CAUSED THE RESULTS I GOT. Which was the original point of the thread. I do have an idea about what could cause it if it were machining in mixed milling mode with lots of short (high resolution) reversals on an axis, but this was not the case. The cut shown was done in conventional mode with each cut pass only having short move reversals on the Z which amazingly only dialed in at .002 backlash before the cut. I have not checked it after because I have been playing with the X & Y for this round of testing. Sure a sloppy machine may jerk and move depending on varying cutting loads, but a consistent loss of position per cut pass in one direction doesn't make sense.
Added:
I might add in order to isolate causes of backlash I turned the screws by hand for some of my tests.