For most common 3 to 4 digit meters, the calibration is done with a dedicated calibration machine like this
http://eu.flukecal.com/products/electrical-calibration/electrical-calibrators/5080a-high-compliance-multi-product-calibra-0
How would they calibrate that one ?
Fluke have a primary source (a Josephson Junction Array) in house. They do, however have to verify that it had not gone wondering so I assume they use a transfer standard (small portable standard that is good for short periods of time) - and they send that around to be compared to other primary standards and make note of any drift. I know NIST also has a Josephson Array based transfer standard (portable) but I assume that is used mostly between countries (but perhaps Fluke and Keysight as well).
Modern DMMs don't drift much. My HP34401a is still within 1 year drift numbers - it was manufactured 20 years ago. So is the Gossen and Agilent DMMs.
That is nice to know, you have all these primary standards in house ?
The are not primary standards - nor standards. They are DMMs.
I purchased a 20 year old HP DMM (made before the Agilent/Keysight name changes). Since the DMM has an internal counter for whenever a calibration adjustment was changed (it is stored in non-volatile memory) - and since it was a low number - it was clear that it was never sent to HP/Agilent/Keysight for their calibration/adjustment service (so the constants in the memory were the same as when it was manufactured).
So I sent it to a local calibration lab to verifying the readings on the DMM (it helped that the service costing about 40$ here so pretty reasonable) .
It was well within the 1 yr maximum drift figures. It is well aged so it is probably going nowhere (unless abused). And once you get the high resolution and high accuracy meter back - you can use it to compare to the other meters (and verify their accuracy) as well.
I do have a few reference voltage sources, a few solid states as well as a Weston Cell. The latter would have been part of a primary standard before the 70's brought the Josephson's Junction Array. But only as part of a group of at least three cells and usually 5-6 cells (so if one drifts off - you'd know which one is broken - if you have one and it drifts - you are in trouble. That is why fungus recommended owning 2 DMMs instead of one. Even if they are of somewhat lesser quality.)