I too will cut a hundreth.
A quick update, I managed to organise the prism holders and carriers, so I had a 0.16mm range in height for the traverse kits. For the prism poles I filed and stoned the high prism holders (two were +0.45mm) so they were all within a range of 0.2mm. The remaining three prism holders were 0.2 to 0.25mm too short, these I placed on our magnet mounts that we attach to structures.
I measured the SX12 instrument height using the level it is 196mm off the tribrach, I don't know where I got it into my head that it was 195.5mm. The Sokkia APS12 traverse kits are adjustable and set to 195.5mm off the tribrach.
I also checked prism offsets, I had a total variance of 0.7mm for the holders I'd selected for the traverse kits and 1.13mm for the holders on prism poles, I might look into improving variance by shimming the holders so they have identical offsets. I have been using prism centring accuracy of 0.5mm in my LSA, but it appears these kits are slightly worse than that.
I checked the pole tip for penetration into a Myzox No.4 Survey Nail, it was 2.8mm.
I also cut slots in some of the prism target boards from the traverse kits and filed them until they were the same height as the prism centre (as viewed from the level). So now I can hang tapes from the prism boards and measure differential levels from resected positions by forced centring. Maybe we'll get second or third order level accuracy, it will tighten up my verticals, while eliminating the need to mark my scan positions on the ground, measure instrument height and perform a level traverse after. This actually speeds up my scanning workflow significantly and I can measure levels while the GLS is scanning.
For control I'll use prism poles with the prism placed close to the ground, check tip penetration and subtract it from rod height. Maybe an easy way to mark the rod tip penetration is to sit it in position and give it a light spray with marker pain.
We had a job last year where the engineers wanted +-2mm accuracy (std dev), it wasn't achievable, so we quoted as accurate as practically achievable. We did get parts of our control network down as low as 3mm @ 95% confidence in 3D, but we also had some higher than that as well, our biggest errors were all vertical, which I put down to small differences in occupation height measurements.
We use the GLS to check the accuracy of point clouds of other scanners used or as a scaffold for registering to.
It's difficult to estimate point cloud accuracy, but if we have prisms attached to the structures we're scanning and we include the scanner's observations in LSA, it goes a long way to providing a good final estimate.