As an alternative to the LM4040, the TL431 is found in many switched-mode PSUs. It's not as good as the LM4040, but most of us have them in our junk boxes. The TL431 is adjustable, but with "REF" connected to "CATHODE", you get 2.5V +/-2% with the worst grade, or 0.5% with the best. Good enough to quickly verify your DMM is OK in the context of the errors discussed earlier, but I wouldn't recommend the TL341 as the core of a DVM reference as the tempco is a bit on the high side.
Another "quick and easy" way to double-check your meters: irrespective of the state of the calibration of a Fluke 73, the reference will be comfortably within 1% of 1.24V. VR1 (an LM385) is to the left of the display.
It's a shame that we probably won't get to the bottom of the AD584 mystery... The small number of problems I've read about prior to this were related to the charging and switching circuitry; if they work at all, they are accurate. So this is a worrying development. As I said earlier, the ICs are recycled, so the completed modules have to be tested before shipping. It's a puzzle
