A good NMR system has hydrogen line Qs of 1e9. To get that, the mag field has to be uniform across the sample to a part per billion. That leads to the arcane craft of magnet shimming.
The superconductive magnets have a few superconductive shim coils, and a mess of (as in 10 to maybe 40) ruum temp shims, each with its own programmable power supply.
If the sample is long and skinny, the field is expressed as a polynomial on Z, the long axis. Z0 is the main field, and there are shim coils to null the higher order terms. I sold a lot of gradient coil drivers, which apply a pulsed Z1 field during the transmit/receive sequances to do some sort of quantum mechanical tricks.
Just the alloy and plating of the antenna coil was a trade secret, to avoid field distortions. Screws were of exotic alloys and the hand selected to be magnetically neutral. Tiny amounts of iron dissolved in a sample were bad. One could buy solvents made of specific isotopes.
*Everything* is magnetic at 1 PPB.I think NMR has been largely replaced by other technologies. It's just a big deal to buy and install and feed a megabuck superconductive magnet.