This is a bit frustrating. I had a 34970a logging my 2dw232 reference since late March to try and extract drift information. Unfortunately the data-logger has limited on-device storage, and truncated the observations to only the last two months of data. I should have verified everything was working and/or periodically offloaded the data, but failed to properly track the project.
The main resistor dividers and current limiting resistor are vishay z201. The reference was placed in a slightly opened carboard box to reduce air circulation, but is not otherwise temperature stabilized. The 2dw232 itself is temp stabilized, with a heater resistor and makes use of the temp-sense feedback voltage from one of the zeners. There are two thermocouples - one loose in the box, and another epoxied to the pcb board.
A plot over time shows the reference voltage and board temperature. The sample interval is 10 minutes, 8333 obs, over 57 days.
A plot using reference V versus the board thermocouple temperature,
Temperature effects dominate and I can't really identify long term directional change - maybe it's increasing.
One point, is that I am not really sure how much of the obvious temperature sensitivity to attribute to the reference and how much to the data-logger which is exposed to fairly wild room temperatures - from around 0C to 20C. Theoretically at least, with the expensive Vishay resistors the 2dw232 ought to be better than the lm399 of the data-logger.
The original intention was to run three different references, so that any variance would be revealed over and above temperature effects.
I will try to post the schematic when I get kicad installed again.