Here are the "before and after" rail level measurements. That is, the results from the epoxy levelling (two curves, one measured going left to right and the other in the opposite direction, just to try to validate the measurement technique as obviously an error at one point will carry right through the remaining measurements). The third, flatter, curve is after shimming. Shims were calculated from the first pair of curves, and were to about the nearest "thou" as this was the thickness of the aluminium foil I used (with some 8 thou shimstock where it could be used). I could have used thinner foil but my wife wouldn't let me eat her Christmas chocolates which had thinner wrappers, even when I said that it was in the name of science and I wouldn't enjoy them...
Click image for larger version.  Name: Rail levels.jpg  Views: 5549  Size: 130.7 KB  ID: 14258
Jonathan - just for you!
To calculate shrinkage, based on my hypothesis that this caused the variation in depth, I had three unknowns. These were S1, S2, and Z. These are the loss in height at the shallow end, the loss in height at the centre (deepest epoxy), and the shrinkage ratio expressed as final height/original "as poured" height. I could measure the depth of the epoxy bed at both points, and the drop in level between them. It was then just a matter of throwing together a set of three simultaneous equations in these variables and solving for each in turn. Plugging the values back in to the original equations gave the right answers, so I am assuming that my maths is OK. Obviously, the underlying assumptions might not be, but that's what we are exploring here.