Luckily Linear and several other companies offer awesome chips with 0.05% initial accuracy on 5V and 10V reference outputs. That gives us a fairly stable calibration source for voltages. Paired with a 0.1% precision resistor, we have a current source and a reference for the resistance ranges of the multimeter. An opamp peak detector opens up AC calibration options.
Currently I use two multimeters, both digital, a Chinese UNI-T VC-840 and a stone age GRUNDIG DM-9 which I hold since many years.
The DM-9 was given to me 20 years ago and has been used on and off. It's a fairly basic meter, no autorange, no fancy stuff (capacitance, frequency). The VC840 on the other side has all of that and a bit more. Some years ago I found the DM-9 in one of my storage boxes and was planning to give it away on eBay or throw it into the trash bin but ended opening it up and cleaning all contacts.
So, more for fun, I calibrated it and was surprised to find that it was accurate to the last digit against my reference. Better even, it remains accurate and has so far not needed a recalibration (despite its probably 30 years of age).
The VC840 on the other side requires recalibration every time as it has drifted off by 2-3%. As both meters are fairly similar in architecture I did some further analysis and found that the quality of the calibration potentiometers and the careful design of the voltage reference in the meter is substantially different. The VC840 uses standard trimmers while the DM9 uses multi-turn trimmers which span a narrow calibration band only.
Lesson learned: It's not always about the features!
Note to self: Need to keep a look out for a decent analog meter.