Tolerances of a couple milimeters? Ha. I hold .005" on bent sheetmetal parts WITHOUT coining it. I can hold .001 if I coin it. Machined/milled tolerances can be held out to six places without much effort and that's using an old 1943 vintage lathe.
Even shipbuilding scale work has tolerances closer than "a couple of millimeters."
I suspect we're looking at tolerances of over a hundred decimal places or more by the time the Galaxy Classes are built.
You're talking about "across a single component." But that's not what we're talking about at all. Think "stack-ups."
One part can hold .001" AT A SPECIFIC FEATURE. You do sheetmetal work... how accurately can you hold the true position of holes on opposite ends of a formed sheetmetal part, relative to each other, if the part is, say, six feet in length?
Now... attach that to another part, also with inherent tolerances... and to another, and to another. Build up a structure made up of THOUSANDS OF PARTS, each of which has a tolerance of its own.
That's called "tolerance stack-up." The more parts you have, the larger the stack-up. So, to make it SIMPLE...
Assume that you have a string of 100 parts, each of which has a tolerance of .001" from mounting feature on one end to mounting feature on the other end. That gives you a tolerance stackup of 100 x .001" or 1/10" across the entire assembly.
THEN, you get into coefficients of thermal expansion... every material grows or shrinks differently than every other material, based upon changes in temperature. So the "tolerances" you have CHANGE... and change from part to part, no less.
Furthermore, you deal with both elastic and plastic deformation of all materials. Just for example, since you brought it up, the hull plating on a naval vessel when first assembled may be of one shape, but after several years in service will have changed shape, sometimes significantly enough to be seen with the naked eye.
So you can hold to .001" on a coined sheetmetal bend... good for you. (I usually expect something on the order of .0003" for a coined feature in an aerospace environment). That does NOT in any way refute the point.
How many MILLIONS of components of various types and materials do you think would make up the "interface" region of a Galaxy-class vessel? That's a MASSIVE "stack-up" isn't it?