I thought it might be fun for someone to take a look on some random numbers.
It's
HP A9 STD-reference, powered by different sources, different times in the year, with uncontrolled ambient conditions (typical +21 to +27c), measured by
two different HP 3458A (calibrated off same source, though) and with
Keithley 7168 JFET nanovolt scanner in between.
This very same reference I used March 2016 to calibrate my 3458A first time.
Original data, taken prior to shipping
Even with all involved, I'm impressed with the peak-peak number and data.
Sample size is non-existent, but if one to judge from this single unit, then theory of having standard A9 3458A reference achieving ~2ppm/year stability might have some real possibility.
If one to count from first measurement after shipment (May 20, 2016),
7.184609 vs last measurement (January 11, 2018),
7.184636 VDC, change is +3.76ppm over 601 days, that is +
2.3 ppm/year. Which is bold statement already, as own meters drift is on similar scale, so it's hard to tell who is who.