More backlash....

I'm getting seriously pigged off by the backlash on the Z axis. I looked at this previously and sort of concluded that the ballscrew backlash was minimal and what backlash I was seeing was due to some form of slop in the quill assembly. 

Certainly, there is the best part of 100um (0.1mm) of lost movement when I jog the Z axis up and down. It's easy to see. The ballscrew responds directly to the jogging movement from the handwheel, with negligible backlash, yet the quill itself shows lost movement. Sounds simple enough. But when I look for radial movement of the quill, I see nothing measurable. The quill end of the ballscrew yoke seems to display the lost movement, yet the ballscrew end doesn't. Can the yoke really be distorting that much?

Nasty looking wear marks on the side of the yoke bearing surface. But not obvious how this could be significant:

I removed the whole ballscrew assembly, examined it, cleaned it up, reassembled it and then torqued it back up to the recommended 30Nm (M8 hex socket cap head bolts). No change. It appears to be behaving as if it is sticking during changes of direction. If the yoke were loose, it would display this kind of behaviour perhaps - but it is far from loose. Or is it? 

Hmm. More investigation required....