One can start creating a disconnect if design tolerances don't match manufacturing tolerances. The uniltaeral tolerincing is common with shafts and holes, but when making parts one typically sets up a bilateral tolerance, which would be 2.2495 +/_ 0.0005 and 2.2515 +/- 0.0005 using the above. The resulting clearance would be 0.002 on average with a tolerance range of +/-0.001, assuming centered normal distributions. Regardless of what the designer wants he really needs to bite the bulltet and to use realistic tolerances in order to produce a robust design. Manufacturing needs to center their process, minimize variation, and constantly work to improve the process.
Some case studies mentioned in tolerancing classes noted that awhile back Toyota had worked to improve their manufacturing tolerances on large assemblies, while Ford just made bigger holes in order to get parts to assemble; guess who had more rattles down the road :^)
Another one compared transmissions, trying to determine why Toyota transmissions lasted longer than Ford units even though both were assembled with parts in spec. Ford had ended up doing a fair amount of sorting to get rid of bad parts, endeding up a distrbution of parts truncated at the spec limits, while Toyota ended up with tighter distributions by keeping theirs centered with less variation. The result was Toyota had assembled more units with parts closer to nominal, so the assembled units perfromed closer to nominal. Wheeler described sorting out bad parts as 'scraping burnt toast'.