There's a couple of these.
One is that the equipment for working metal needs to be remarkably sturdy and stiff. This is why a lot of those "build your own metal lathe with some angle iron and a drill motor" aren't really going to hack it. And why there is such a difference in cost between a desktop CNC with a laser or (printing) hot end, and one that can carve metal. The machines are heavy, built of steel. The tool holders and other parts have massive bolts and multiple set-screws, and when you lock a setting in, it takes brute muscle power to tighten all the bolts and clamps.
And yet; so much about why we approach a project in a certain way, why cuts are made in one direction over another, why the trade-off between speed and tolerance, have to do with the fact that everything flexes anyhow.
Perhaps related to this, the core calculations for machining -- and engineering -- are at the heart pretty basic physics and geometry. And it is an essential skill to be able to work like this, practically from first principles, to figure out the cross-section of a load-bearing member or the weight of a finished casting.
And yet, against just like in engineering, the devil is in the details. All those places where the abstract form just doesn't carry sufficient information, where the first-order effects are edged on to by second and third-order effects. You really do need to model the thing you mean to do...but you also need to know that the model lies.
Probably the least paradoxical, is that machining is done to tolerances of tens -- ten-thousandths of an inch (I'm barely up to accuracy of 0.0005 myself). But we don't work within arbitrary decimal numbers out to those five places...instead the vast majority of work is done to the nearest appropriate "round" number, or one of the stock fractions that come up over and over again. So I could lathe something to 0.25000", and I might even set the dials to that value, but basically I'm lathing to 1/4" in diameter.
This comes from three different directions. One is that hardware, tools, and stocks are available only in certain sizes -- many of them simple reduced fractions. Another is that as tight as the tolerances might be in one place, there is expected slop all over; necessary gaps to allow parts to move, the known tolerance of manufactured hardware or raw stock, etc. So when you hit a gap or other fit-up, you take the chance to jump up to the nearest simple fraction...instead of blindly adding the exact calculated clearance at each step and working outwards in numbers that are increasingly random-sounding. The last is because it is a heck of a lot easier to remember, and to calculate, and to do the in-your-head approximations for measurements you've dealt with many times in the past.