Thanks Joe, for a very erudite analysis and refresher on the MV106 specs.
Of concern here is the stability of the SVR, in this case compared to the MV106. If the MV106 is not good enough as a comparison, then clearly its inferior to the SVR, and that can easily be demonstrated, QED: compare outputs of both units over time and compare it to a another reference, a calibrated DMM.
The MV106 reference is a 6V zener IN829? that has the ~ tempco of 5ppm/
oC as the AD587. Its uses a chopper amp with some gain and a non-KV resistance network to scale 10.000 00V, 1.000 000 V and 10.000 00mV, and a maximum resolution of 10nV. Accuracy and stability wises, it is at least on paper <= to the SVR, as you describe.
The dividers of the MV106 are all tied to the performance of the IN829 and despite what can go wrong in a discrete design, dial down to 1.0 uV on the 10mV scale and it will hold this reading for at least a month, to ~ +/-2ppm.
You can dial output in 100nV increments, the limit of the 3456a resolution in the 10mV scale, and it will hold for hour.
All the above regardless of weather.
OTAH, tie my SVR to a quality KV divider, even suitably buffered, and you'll be lucky to get 1uV out of it, that's of course out of its spec, 1uV being 0.1ppm.
Getting it checked by you and getting another would certainly shed more light on this weather or malfunction issue.
I suspect key to the MV106 performance is that the now well aged zener is fed by a very well regulated constant current source in addition to the chopper amp. The SVR is at the mercy of line and load regulation which is considerable at 100uV/V, which can explain a lot of its variation. By itself, the IN829 is far inferior to the AD587, so it would be interesting to build the stability circuits of the MV106 around the AD587.
Barometric effects are known, albeit it doesn't happen uniformly and varies in magnitude, from none to some. Whether this is my phenomena or something else related to weather [ such as affecting the power feeding the SVR, or changing humidity] remains to be seen. Suffice to say, the SVR reads best when the weather is best.
http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=852292&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D852292Rather than calculate a correction coefficient, which is possible once enough data is obtained, I simply make a critical measurement only when the conditions are identical to the original calibration, and saves the trouble.
On the subject of physical issues affecting voltage references, I've never heard of changes with physical position, as suggested in volt-nuts:
http://www.febo.com/pipermail/volt-nuts/2010-September/000465.htmlIts the only report I know of this effect, and Dr. Stellmach is reputable. So, you never know what new things can occur.
>>The MV106 is spec'd to < 5uV stability [ I get it practically to 1uV or less under controlled environmental conditions] and resolution to <= 100nV
We have several MV106's too, including one with the built in null meter. ...
I think the problem might be a combination of units, test setup, and it could be a very broken board too.