The spy sats bought a bunch of ball bearings (these might be a foot in diameter and are speced to be extremely low noise at low turn rates). They tested them all (using a phono needle resting on the outside of the bearing while it was slowly turned). The ones that made the least noise went in the sat while the others were sealed in a plastic bag and put on a shelf in the clean room.
I was told that when Hubble came along, the US no longer had the capability to make those (I'm not sure if that was true). In any event the ones that went in Hubble were the least noisy of the ones that had sat on the shelf. My summer job was (largely) testing to see which was the best. A cool job.
It turned out that a normal distribution centered at 5 ohms was not quite right. It was a normal distribution, but with a deep notch taken out right at 5 ohms. All the resistors that tested to very close tolerance had been bagged separately, and sold at a higher price.
(The context was why you might want to put, say, four 5-ohn resistors in series, rather than just use one nominal 20-ohm resistor.)
In the old days, it might have been cheaper than the more accurate resistors. These days, high precision resistors are relatively inexpensive.
But of course it's better when the tolerance doesn't matter (or only of a few components)