I'm having issues with getting within tolerance on a project which have two linear rails. After spending loads of time on trying to get within tolerance i'm starting to think that i might be reading the documentation wrong and/or misunderstanding GD&T. So here i am, asking you guys for a sanity check before i go deeper into this rabit hole...

The setup: I have two Hiwin RG35 linear rails spaced 400mm apart. Its these guys: https://www.hiwin.de/en/Products/Lin...0H/p/12-000287
If one download the catalogue, it specifies the required properties for the mounting surfaces like this:
Click image for larger version. 

Name:	RG35 rails2.PNG 
Views:	0 
Size:	47.9 KB 
ID:	481238
All reference surfaces have been milled and now also ground...and i'm still way of...maybe...

So the parallelism of the two shoulders P must be below 0,01mm according to the datasheet . If i mount one rail and put a dial on the oppesite shoulder i'm somewhere below 0,002mm worst-case over the entire length, so well within spec on the shoulder! But if i put the same indicator on the mountingsurface i'm hitting almost 0,04mm worst-case over the travel length. So i've been chasing this and ended up grinding the surface... i have a mirror finish and zero(!) flaws if i blue it up using a high precision granite "straightedge"... i can't do better than this and i'm still 4x of the tolerance... this can't be right! And now i'm starting to think that i'm misunderstanding these tolerances completely....

Since the rail is only 35mm wide, even the smallest imperfections a magnified over the 400mm rail-distance with a factor of almost 12x. So me having 0,04mm on my dial from one rail to another tells me that i'm +-1,5µm off at one point below one of the rails....this seems unrealistic for me to correct...

Is there any way i can verify that i am within tolerance on this using the rails themselfs? I don't have a surfaceplate nor a CMM large enough and i'm not even sure that it will account for the "twist" i'm seeing....