If you find this stuff interesting, pick up a book aimed at hobbyist metal lathes. Just reading a little about how you set up a tiny home version can give you a whole new level of understanding and respect for this kind of stuff. Structural components are different but in that era and even to some extent today, most precision parts were machined with two primary tools: the lathe and mill. Today both would be CNC and you'd add other modern tools but the CNC lathe is still a lathe and they work the metal in the same fundamental way.

Due to the dispersed mass production in WW2 it's probably not totally true that they wouldn't fly without a custom tooling setup (it'd be a maintenance nightmare if two examples of the same model from different plants diverged dramatically, altho of course it happened) but you can be certain that it's of critical importance.

It's because the designs weren't made for mass production. The given tolerances were nominal and you were supposed to hand-fit parts together.

For mass production, they had to be re-tooled with larger tolerances which also allowed for interchangeable parts.

That's an interesting though, like software documentation that hasn't been updated over years of patches.

Do you remember where you read that? Sounds like an interesting bit of history.

A long write-up of an equivalent for the F-1 engine is here: https://arstechnica.com/science/2013/04/how-nasa-brought-the...

The analogy to out-of-date software documentation is very apt. As problems or inefficiencies popped up in the factory, they'd change things and just not write them down (because of deadline pressure, often). If you wanted to build new ones you'd have to either get the factory workers to show you how the thing was actually built, or tear down a working example to reverse engineer it. (Which has been done!)

Also sounds like a git repository where people were working on local branches to do production work and then when they left the company, never pushed their branch upstream.

That's assuming there is a git repo or any kind of version control at all.

They're not assuming. They're analogizing.

You might have fitting issues with what I like to call "slide ruler error".

How things came to be and are, are often at times not documented. Dimension and tolerancing is lost over time too.

Statistical quality control was not widespread at the time outside of a few enlightened industries. The Army was much more likely to buy a lot of parts where they could expect most of them would fit, rather than having all of the parts in statistical control within the tolerance limits.

A huge amount of industry all over the place ends up like this. The engineers produce the theoretical form of the design, and the tooling up process refines it into a practicable-at-scale reality.

I only worked tangentially in the field, but I get the feeling that modern semiconductor manufacturing is sort of the same, except with more excel sheets and the like?

Semiconductor manufacturing is essentially the interchangeable parts engineering taken to the extreme. The tolerances are so large (relative to the feature size) that instead of trying to eliminate them you exploit the relations between various manufacturing tolerances such that the errors cancel out.