I have followed this thread wondering how it might play out. Just for reference, I live in a world of hole location, also referred to as true position, and the relative fit of components, sometimes referred to as class of fit. My experience is with what most of you would think of as gearboxes, differential carrier mountings to be exact. Think of the data Rob presented but in a three dimensional cube as opposed to a flat, two dimensional MFT plate.
A quick look through the dimensional layout of the tops indicated a worst case true position of about .0075". True position is double the actual distance a hole deviates from perfection. It is not uncommon for the spacing of bearing bores for gears to be toleranced at .002 (fairly tight) to .005 (pretty open in general, but depending on the application can be more difficult to accomplish than .002 in a different place). For an MDF top to be bored at the accuracy documented is pretty impressive, and most of the hole were on the order of .004 or less. In addition, hole size over the three tops varied by only .003 which would add another .003 to true position variability, but only when using the smallest hole along with the largest hole. When looking at any individual top the hole sizes were closer than .003 within each given plate, meaning the variability would be even less. BTW, .003 hole size variability over the time span represented in the samples is not too shabby either. I am estimating, off the cuff, that Festool is using about 50 percent of the tolerance they have allotted themselves. We are getting a very good platform on which to do our work.
That said, I agree wholeheartedly with Rob's assertion that more variability in position for squareness or 45 degree cutting will come from the hole positions than from the size of the dogs. IIRC, Rob said they are using a tolerance of .002 for size which would only account for .002 position variability (when loaded against the side of a hole), much less than described above. In addition, I am betting LV is holding better than .002 especially within the dogs that wind up in the same kit. It is probably more like .0005 but Rob is not going to make this claim on the off chance it doesn't happen.
Final observation: we use 20mm bolts by the hundreds of thousands. I have a handful that I use for dogs. They are a perfect fit in my MFT/3 but will not go into my older MFT 1080 unless I tap them in with a mallet, nor will they go into a Walko without the mallet. If I were to dust the diameters for the smaller holes they would 'seem' too loose in the larger MFT/3 holes but it would only be .003 to get there. Three thousands seems really sloppy until you work thru the numbers. Rob did it for us.
The takeaway for me is that we have 'assumed' perfection in the top, and that all our 'problems' will be from loose fitting dogs. Quite the contrary. The fit of the dogs will contribute less to accuracy loss than any other factor if you load the force against the wall of the hole, which should be a default technique if you want to be assured your stock, or fence, is also tight to the reference surfaces.
