Thank you for your answer.
Just to clarify:
1. Inverting Pan and Tilt is "bad", it can mess with slow movements
2. It has nothing to to with the interface or the way packets may be split up
That said, I had tried already moving the channels manually; moving the fine channel slowly back & forth from 0-255 results in a good movement of the head.
Moving the coarse channel also works ok but of course much more jerky due to the resolution.
Your idea is interesting; I understand you are saying the fine channel "overshoots" the coarse channel; how can I change from 255 to 0 and the coarse at the same time from, say 0 to 1 at the same time to observe this? If I keep the fine at 255 and move the coarse to 1, it will move the 1,255, which is of course in the same direction, but we want to test 0,255 to 1,0.
In either case, how can this be implemented in the DDF? Is there an undocumented parameter for the finedmxchannel that says range=224 or something? I suppose we should keep this a binary value?
It is also worth noting that this isn't completely random; certain segments of an ellipse are predictably bad and others predictably smoother. I other words, I believe to have observed the jerkiness to be deterministic. This is why I am suspecting a programming issue.
Fwiw I have also now purchased a USB to DMX to test another interface, and I have also tried Ethernet vs Wifi and wireless DMX vs cabled in all combinations, all with the same result.