Fluke 5440A calibration

[Rax]

Well, maybe the extended "bake-out" bdunham7 is what was needed here?...

After being on for about two weeks now, and creeping ever so slightly to perfect agreement with the Prema (from about 5ppm after the first INT CAL, to then 3ppm), it's now at just about perfect agreement.

I did one more INT CAL this morning - may be the 4th or maybe 5th - and then I saw it coming essentially dead onto Prema's 10V (within a measured +.09ppm and -.05ppm). Overall, today I saw these infinitesimally small variations around 10.0000000V. Far less than .1ppm for all of the time. Less than .05ppm from 10V actually, most of the time, on either side of 10.0000000V.

I'll keep on watching it for the next few days or maybe weeks, but the trend seems very positive.

[Rax]

I wonder if anyone's aware of a way to output the cal constants generated during the last INT CAL. Either displayed on the screen, or saved in a way that can be extracted from the instrument. I think they are being automatically printed (or can be) when the procedure is being completed. Also, supposedly, they can also be displayed on the screen (though I can't find any indication how that's supposed to be performed). Until I'd run the INT CAL again, it seems reasonable to believe the last generated cal constants should be available to be displayed.

Given how close the measurements are currently, I'd want to have those saved somewhere, in case they need to be resurrected somehow.

[Dr. Frank]

The Cal Constants can be printed via serial interface. Print is accessible via instrument Menu.
That's they way how I calculated the calibration shifts of my 5442A.
Did not figure out yet, if there's away to access via IEEE
Frank

[maat]

The Calibration constants can be extracted via GPIB. Do you have a GPIB adapter and know you way around in Python? If you are using Linux, you may directly use my Fluke5440B libarary: https://pypi.org/project/fluke5440b-async/ and see https://github.com/PatrickBaus/pyAsyncFluke5440B/blob/master/examples/test_functions.py for a full example.

Otherwise, check out the GCAL GPIB command and my implementation of extracting the parameters here: https://github.com/PatrickBaus/pyAsyncFluke5440B/blob/64f4869891e0b51c1e66b73948303a7bed34eab8/fluke5440b_async/fluke_5440b.py#L974

[Rax]

After some fat-finger fumbling with the Prema - such a half-baked software/menu system...  how does one exit, or cancel out of unintended modes on PROG/CONT?... - I've decided to redo INT CAL (after all, it seems to be recommended as a daily procedure per SM) and I'm a bit less dead on at around .7-.6ppm off of a perfect 10V on the Prema. My plan was to somehow record those "near perfect" INT CAL-generated cal constants, but I don't currently have a way to read GPIB. I do have a kit of parts for the "USB GPIB" but never got around to putting it together.

If anything, I think this solidifies its closeness to either absolute/nominal 10V, or at least to Prema's 10V. I'm still looking for input on whether all this stuff I'm seeing is expected behavior from a fully working but recently resurrected unit. I'm inclined to think it is.

Also note the Prema was re-zeroed, both in auto and 20V manual mode. To snap it our of its bonkers programming place (I have no idea where it entered, but I couldn't figure a way out), I also had to restart the Prema once. I keep it permanently on as I see not reason for it to not be perpetually warmed up.

[Rax]

The Cal Constants can be printed via serial interface. Print is accessible via instrument Menu.
That's they way how I calculated the calibration shifts of my 5442A.
Frank

Frank - can you please share more on how exactly you're reading that? What sort of cable, how do you connect it to a computer, etc? I assume you don't actually use a serial port printer (which I think is how they intended it).

[RandallMcRee]

I wonder if anyone's aware of a way to output the cal constants generated during the last INT CAL. Either displayed on the screen, or saved in a way that can be extracted from the instrument. I think they are being automatically printed (or can be) when the procedure is being completed. Also, supposedly, they can also be displayed on the screen (though I can't find any indication how that's supposed to be performed). Until I'd run the INT CAL again, it seems reasonable to believe the last generated cal constants should be available to be displayed.

Given how close the measurements are currently, I'd want to have those saved somewhere, in case they need to be resurrected somehow.

Page 3-5 of the operators manual tells you how to display on the screen.  Give it a try and just write them down!  Also you should probably start logging the temperature to make sure your 5440 output is as stable as claimed around that magic 10 C band. 

[Rax]

Page 3-5 of the operators manual tells you how to display on the screen.  Give it a try and just write them down! 

Thank you, Randall, that's it! And I can't even invoke the differences between the "5440 series" Operating Manual vs the "5440A" specific one (though the latter has it at p.3-9), 'cause... the latter has it at p.3-9! Thank you for going into it and digging that up. Very helpful.

Also you should probably start logging the temperature to make sure your 5440 output is as stable as claimed around that magic 10 C band.

Goop point, and I always do! Truth be told, given I haven't completely closed this up (no top and bottom cover yet on, though I'm doing my best to direct design airflows through its innards, and shields are on, etc.), the ambient air temperature and airflows may have a higher than designed impact on its stability... One could tell via GBIP if the garage door was opened to get the car in by my better half!

In any case, I've taken notes of its - apparently - 20 calibration constants and given it's displaying 10.0000000V right now on the whole 8 1/2 glory of the Prema, I think I'll call it a fruitful day!

[Rax]

logging the temperature to make sure your 5440 output is as stable as claimed around that magic 10 C band.

Thinking a little more about this, Randall - it's likely why I'm seeing these excursions on the low end of a reading on my meter for the 5440A output. It seems that typically, a warmer environment would trigger lower readings for a certain voltage output (at least with my units here), which may be counter-intuitive for most (including me). So if I, say, see 28C at my bench (hot for all lab typical conditions), I'll likely see a "lower than 10V" reading on the Prema (for instance, 9.9999995V, therefore lower than 10.0000000V). Importantly, I should probably suspect the Prema for this tempco far more than the 5440A, with its ovens for all critical stuff. The Prema has the famously uber-stable LTZ1000, but none of its circuitry is temperature stabilized. Almost everything of substance on the 5440A is.

Stepping back, there's all sorts of thermal inertias of things from heavy walls to slabs to... instruments to air convection and whatnot, so it's a bit hard to really grasp how this plays into a metrology reading, but my background in both physics and building engineering (HVAC etc.) helps. I mention it because that's what I think needs to be considered.

[Dr. Frank]

The Cal Constants can be printed via serial interface. Print is accessible via instrument Menu.
That's they way how I calculated the calibration shifts of my 5442A.
Frank

Frank - can you please share more on how exactly you're reading that? What sort of cable, how do you connect it to a computer, etc? I assume you don't actually use a serial port printer (which I think is how they intended it).

I have an older OKI LASER printer which still features an RS232 interface, so you just need a serial cable which fits both sides. I don't know the exact configuration of this cable, i.e. if it's got twisted TX / RX lines, handshake lines, etc. With RS232 you always have to try out, or read that d...ed manual.
On the front panel, there is a menu to print all cal constants. Again, the operations manual is your friend.
I'm adding a photo of the different snippets here, documenting the calibration progress of my unit.

Frank

[Dr. Frank]

logging the temperature to make sure your 5440 output is as stable as claimed around that magic 10 C band.

Thinking a little more about this, Randall - it's likely why I'm seeing these excursions on the low end of a reading on my meter for the 5440A output. It seems that typically, a warmer environment would trigger lower readings for a certain voltage output (at least with my units here), which may be counter-intuitive for most (including me). So if I, say, see 28C at my bench (hot for all lab typical conditions), I'll likely see a "lower than 10V" reading on the Prema (for instance, 9.9999995V, therefore lower than 10.0000000V). Importantly, I should probably suspect the Prema for this tempco far more than the 5440A, with its ovens for all critical stuff. The Prema has the famously uber-stable LTZ1000, but none of its circuitry is temperature stabilized. Almost everything of substance on the 5440A is.

Stepping back, there's all sorts of thermal inertias of things from heavy walls to slabs to... instruments to air convection and whatnot, so it's a bit hard to really grasp how this plays into a metrology reading, but my background in both physics and building engineering (HVAC etc.) helps. I mention it because that's what I think needs to be considered.

Well, I've got a basement lab which features ultra stable room temperature all over the year, i.e. basically +/- 1°C.
You can have constant temperature over days and weeks within +/- 0.2°C, w/o any HVAC, only heating in winter time.

Unstable room temperature, i.e. variations > 0.5°C over the measurement run (even with ACAL instruments) will ruin your day.
Variations of more than +/-2°C make precise/stable absolute measurements nearly impossible. Transfer / difference measurements are another story.
 
The 5440 might be more stable with his ovens, but my hp3458A has a (typical) T.C. w/o ACAL of +0.5ppm/°C, which you directly see in your measurements.
My better half once opened the small basement windows to get fresh air inside, when my 24h stability measurement was running.
That -0.5°C r.t. change created a -0.3ppm dip, so I was able to calculate the T.C., which comprises of the T.C. from the LTZ1000A circuit, plus the T.C. of U180, which can be separated by monitoring CAL? 72 and sketching that over TEMP?

The heating of the room by the instruments (200W in total, I guess), plus the 36W neon lamp, plus the 100W from the operator therefore pose a real problem for high stability and transfer measurements. You first need to let the room temperature stabilize to the new level, and then you have to make your transfers quickly and monitor the temperature to make sure it doesn't change more than 0.2 .. 0.5°C.

I append the temperature log from last month.

At the beginning, the data logging stick had to cool down from the temperature in my office upstairs, @ over 25°C, where I always read out the data.
Then you see weeks of mostly constant temperature. Outside it was quite warm, over 30°C, I think.
In the 2nd half of September, the outside temperature went down below 20°C. I have not yet switched on the gas heating, which stabilizes the r.t. of the lab even more.
On October 1st, I made my regular measurements, where you clearly see the warm-up by the instruments in the night before, starting @22:30.
Next morning ~ 8:15 ACAL ALL with fluorescent lamp on, 7x Ohm comparisons 8:42 .. 9:00, 3458A TEMP? increase by 0.1°C.
A break-fast later, there's the comparison of 13 volt references between 9:38 .. 10:06. Temperature increase inside the 3458A was +0.1°C only, but R.T. changed by +0.3°C already.
ACAL temperature had been 0.6°C lower, but the absolute reading of the 3458A does not interest here, only the transfer between the 13 sources.

I recently had sent two of my references to a fellow volt-nuts, together with another temperature logger.
I got very similar temperature profiles from his lab, and from his measurement session.

It would be helpful to make remotely controlled measurements, no lighting and no operator in the lab, but then you'd need a precision scanner.

Frank 

[Rax]

Thank you, Frank, for sharing that data.

In my case, I'm relatively lucky to be located in Southern California - BTW, for some mysterious reasons I have some possible explanations for, there's almost no basements here!...  - which may have some of the steadiest, mildest weather on earth. This year's been extreme by our standards, so it's been unusually difficult in the lab - and who knows what the future holds - but usually I'd be well within a very modest variation (maybe 2-3C tops during a certain cycle, like a day or even a week). It does tend to be rather hot for a mysterious reason (it's on a concrete slab with two fire rated, exposed masonry walls on each side and a garage door that's receiving a fair amount of sun in the afternoon; there's a master bedroom above, that's it, and somehow it's by far the warmest space in the house!). But I have a heat pump in there, with both cool and heat cycles, so I can shave off the worst of spikes. The car does need to go in and out, so far from ideal. Home metrology! 

[Rax]

As I'll continue evaluating the state of both accuracy and stability of the unit, one other possible action item along the lines of bringing it back to spec is adjusting the power supplies. Thoughts?...

I'm of two minds on this, on one hand I really like to see power supplies readjusted to specifications, on the other I'd hate to fix what's not broken and maybe create some problems.

[bdunham7]

I'm of two minds on this, on one hand I really like to see power supplies readjusted to specifications, on the other I'd hate to fix what's not broken and maybe create some problems.

Are they out of specification now?  What is your mains voltage measured inside the unit (if you can measure that reasonably safely)?

[alm]

I'm of two minds on this, on one hand I really like to see power supplies readjusted to specifications, on the other I'd hate to fix what's not broken and maybe create some problems.

In general, if a power rail is within tolerance, like 11.5-12.5V, then adjusting it to the middle of the tolerance band won't make a circuit work any better and may require adjusting other parts of the circuit. If it's outside specifications, then I would first look into a component defect like a dry electrolytic or drifted resistor before adjusting.

[Rax]

Thank you both - bdunham7 and alm.

In a nutshell, some of the rails were not drastically off, but definitely out of spec last I checked. Some were spec'd very tightly, and that's how I mean out of spec. I'm not sure why some would need to be like "5.15V" (from memory)... . [later: (+-0.00V?... really?!)] I'd need to dig out my readings from some bottom of a pile of handwritten notes [see just above].

That said, I fixed a busted rail and recapped the entire unit (with some small exceptions) since - aside from just making it work. So it's possible whatever stressed parts were there before - caps, though I'll still be suspicious of drifted or just stressed carbon comp resistors - they've been replaced.

I'm pretty reluctant to mess with it very much from this point. But may take a hard look as some of the carbon comp power resistors, which I regard as just unfit for the kind of service they're in. Sign of the times to employ them.
__________________________________________________________________________________

Actually, I found my notes. Mind you, at the time I was still using the "5440 series" SM. I'll have to cross check to see if all applicable to the 5440A, but I don't expect many surprises. [later: no surprises]

As the notes say, this is from back in April, when I was just embarking on this behemoth of a project. I have not gone back in there and rechecked the rails.

[Rax]

Not sure if anyone has any fresh thoughts on the PS measurements. Both bdunham7 and alm made very pertinent comments, I think. I'll have to evaluate my odds (and star alignments) before poking on the rear side for the line voltage as it comes in. There's more truth to it than most of y'all know, there's some fairly long extending cords providing power to my gear. Garage setup all the way.

Tonight though, following a fresh INT CAL for the 5440A and a fresh 20V zero for the Prema 6048, they got very quickly within less of .1ppm of 10V (and each other, of course). Most evening they hanged out together somewhere less than .07ppm of perfect agreement on 10V. I didn't quite get 10.0000000V (got some 10.0000002Vs), but that's like winning the lottery. But I confess being just captivated by seeing that many zeroes!

[Rax]

Long(er) term observations leave me pretty floored.

Given relatively fresh INT CAL of the 5440A and zeroing of the 6048, their agreement is just about as high as ALL of their 8 1/2 digits allow them to agree. Literally, most of the time they agree on all digits (and be on 9.999999V or 10.000000V). If I see variations around that, they'll be typically within less than .1ppm of 10V (sometimes the same reading over multiple 20s integrations by the 6048). If things "slide down a bit" (wilder variations of temperature, no recent INT CAL, etc.), the highest divergence I would see may be as high as .5ppm - but this is very, very rare, and easily corrected by cursory operations as described.

I wanted to make sure I accrue some statistical relevance for this data over some time period and temps, and I can pretty much "provisionally conclude" based upon a few weeks the agreement on an "8 1/2 digit worth" accuracy on 10V (factor in here my "932B level stability" 931B which is agreeing to sub-ppm levels too with these two) is solid.

As branadic and Dr Frank argued, the 5440A is, in (now, also my) opinion as accurate as it can possibly be. I think it's about as accurate as when it last left the Fluke cal facility in Everett WA. I definitely don't think any calibration is needed (at least at 10V; other outputs rely on linearity of the instrument as Dr Frank stated and so they're probably best left alone). I feel this also vets my Prema and the 931B. I am happy to have a few solid references (at least I feel I can qualify them as such).

[alm]

While seeing all nines or zeros is certainly satisfying, I don't see it as anything but coincidence. It's like saying you and another person are driving the exact same speed on the highway: not something I'd adjust a speedometer to.

The way I look at it, the P6048 was compared to the traveling reference X days ago. This reference's value was known within u1 ppm. After X days, the P6048 drift will be within u2. The uncertainty due to tempco of the P6048 based on the difference in temperature when measuring the reference X days ago is u3. The noise in your measurement (standard deviation over samples) is u4. The uncertainty due to non-linearity (depending on how different the voltage of the reference and 5440B) is u5. Then add up all type A and type B uncertainties as standard errors per GUM and multiply by coverage factor, and the result is a statement like "The P6048 at this point in time and temperature will measure 10V as 9.99995V +/ 5 uV/V (k=2), and based on that the 5440B's output when set to 10V is 9.99995V +/ 5.5 uV/V (k=2)".

The 5440B is certainly very stable and linear. So if it were 100 uV/V out, I'd suspect a defect. And the fact that it's quite close gives confidence that it's working properly (in that range). But that doesn't mean it's within 0.1 uV/V either. And I wouldn't want to run it 24/7 for most predictable drift given the noise and power usage

[Rax]

While seeing all nines or zeros is certainly satisfying, I don't see it as anything but coincidence. It's like saying you and another person are driving the exact same speed on the highway: not something I'd adjust a speedometer to.

I [consider I] understand your point, but I think ultimately, this is not unlike statistics as an epistemological approach in physics - as much as it can do all of the things physics is expected to do (predict phenomena which can then be verified in experiments to check the validity of a theory based on statistical thought), it still essentially doesn't have ontological substance (it doesn't positively say an individual event will take a certain course). Differently put, just because metrology is based upon thought articulated around uncertainties, it doesn't mean one's volt cannot objectively be accurate. It just doesn't. I'd definitely adjust my volt around conclusions such as the above, and expect it'd pass calibration checks against traceable standards.

What I posit above is:

• the accuracy of both the 6048 and the 5440A (and the 931B 731B and the FX) is very high (at 10V!)
• their aging is infinitesimally small at this point (to borrow a term from differential calculus, which was created for no other reason than deal with physical phenomena such as this) and it's been so since their last "calibration with adjustment" (this is intended to address the comment on highway driving)

Also differently put, the fact that a reference (FX) adjusted decades after these two above agrees with them as closely as it does (my own measurements of the FX) points to the same. The likelihood of all these references across years of adjustments and checks to be wrong by the same infinitesimally small amount diminishes the likelihood they're all wrong (a statistical formalism statement) by an amount proportional to the size of that differential (smaller differential = higher likelihood of accuracy). Therefore, it's likely that they're right (I also posit the two are logically exclusive).

It's why voltnuts accrue multiple 932Bs 731Bs at their bench - to diminish their divergence from the objective volt by averaging, evaluating (collective) aging, etc. Diminish the likelihood of them being wrong.

[Rax]

I wouldn't want to run it 24/7 for most predictable drift given the noise and power usage

I'm building the guts to turn them off.... I promise.

[Rax]

I wouldn't want to run it 24/7

The actual (= based on reason) motivation for having all these on for so long is accruing the hundreds of hours of observation I have piled up with them, in various combinations of measurements and configurations. My WFM situation (I am currently still hybrid), and the fact that WFM occurs at the same desk which is my bench, allows me to monitor these things for many hours at a time (while not yet having the capacity to automate this empirical data accrual).

[alm]

I [consider I] understand your point, but I think ultimately, this is not unlike statistics as an epistemological approach in physics - as much as it can do all of the things physics is expected to do (predict phenomena which can then be verified in experiments to check the validity of a theory based on statistical thought), it still essentially doesn't have ontological substance (it doesn't positively say an individual event will take a certain course). Differently put, just because metrology is based upon thought articulated around uncertainties, it doesn't mean one's volt cannot objectively be accurate. It just doesn't. I'd definitely adjust my volt around conclusions such as the above, and expect it'd pass calibration checks against traceable standards.

There are two reasons why we need statistics in metrology:

• Any measurement you do, with the exception of some fundamental measurements like counting atoms, has some variability associated with it. If you measure a block of jelly ten times with a tape measure, you'll likely get ten different outcomes based on the compression of the jelly, the angle at which your eye looks at the tape measure, how hard you pull on the tape measure, etc. If you have ten different people measure ten times, you'll get even more variation. We use statistics to describe such processes.
• We can not predict exactly what all atoms in a voltage reference or resistance standard are going to do. Or how every standard is going to change over time, temperature, air pressure etc. So what the manufacturer does is based on their own testing and their process control, they will specify a number that if you measure 100 standards, at least 95 will stay within. In general they will guard band this number so their process control doesn't have to be so tight. This is the number you will see as specifications. They might say an instrument will not drift more than 10 uV/V/year.

The alternative for modeling 1 using statistics is saying you have a perfect model that can predict all the noise and other contributions to the variation in measurement. I would love how you built such model. Regarding #2, you are right to say this is describing a population, not a single instrument. Characterizing an individual instrument's performance like short term and long term drift, tempco, etc is certainly viable. Usually it means having to acquire a sizable volume of data against an independent, much more stable reference. For example, you could measure the drift of a 3.5 digit meter against your P6048 by measuring it now, again in 1 year, and seeing how much the 3.5 digit meter changed. The drift of the P6048 is likely to be negligible compared to the 3.5 digit DMM. Or you could have your voltage references calibrated against a much more stable reference (like the Fluke reference bench that's regularly compared to their Josephson voltage standard. See this Fluke article on how well they can estimate the drift of their voltage references after a series of calibrations. But stupid Fluke still uses uncertainties for these estimates. What data do you base your "better" estimates on?

their aging is infinitesimally small at this point (to borrow a term from differential calculus, which was created for no other reason than deal with physical phenomena such as this) and it's been so since their last "calibration with adjustment" (this is intended to address the comment on highway driving)

Aging is a motion. Since it's hard to measure the velocity since we're talking about very small velocities, one generally measures it based on multiple points over time. Ideally spaced a fair time apart (months at least) to reduce the requirement on how accurate you need to measure the difference. What are your multiple points?

In metrology, you normally would ask a cal lab to never adjust, so you can track the drift yourself. Do you have access to a previous cal certificate that says what the measured value was and with what kind of uncertainty? Even cal labs have uncertainties, and when calibrating high-accuracy instruments their uncertainties may not be that much better than the meter. Even when it was adjusted, the cal tech has no interest in putting it at exactly the nominal value. Any value within the specified uncertainty will do, and no adjustments will be made as long as it's within specifications. So all you can say as that during the last calibration, it was likely in specifications. Hopefully 24h, but maybe 1 year, depending on how good the cal lab was and what the customer asked for. Add to this that a DMM or calibrator is a complicated instruments with a lot of components that can affect aging, and you can't assume them to drift monotonically, never mind linearly like the Fluke 732B voltage reference in the article above. There will be multiple components with their own drift rates and time constants. Think multiple springs with different spring constants in series. Based on this, I'd say any statement about drift of the P6048 (that you compared to the FX reference once) and 5440 (that you compared to the P6048 over a short stretch of time) is pure fantasy.

Also differently put, the fact that a reference (FX) adjusted decades after these two above agrees with them as closely as it does (my own measurements of the FX) points to the same. The likelihood of all these references across years of adjustments and checks to be wrong by the same infinitesimally small amount diminishes the likelihood they're all wrong (a statistical formalism statement) by an amount proportional to the size of that differential (smaller differential = higher likelihood of accuracy). Therefore, it's likely that they're right (I also posit the two are logically exclusive).

While it may be plausible, you just can't make any definitive statements based on that. Certainly not at the sub-ppm level. You're operating at the edge of what's physically possible with zener references. If you had a series of comparisons against a Josephson voltage standard (see Fluke article on predicting drift of voltage references, which are much easier to predict for reasons I just mentioend), then it would be a different matter. But you really can't claim sub-ppm uncertainties based on some hand-waving.

It's why voltnuts accrue multiple 932Bs at their bench - to diminish their divergence from the objective volt by averaging, evaluating (collective) aging, etc. Diminish the likelihood of them being wrong.

The 932B, is that a newer, unreleased version of the Fluke 931B RMS differential voltmeter? While averaging multiple references can certainly reduce drift, since the variance of the average of N independent normal distributions is s^2/N, the references are never truly independent. Standards labs like the NBS (now NIST) used to maintain quantities like Volt and Ohm using a pool of artifacts, like Weston Cells or standard resistors, that were very carefully monitored and averaged. They used two different kinds of Weston cells to balance out systematic factors. Yet when more stable quantum standards were developed, it was found that both the NBS Volt and NBS Ohm were drifting relative to the quantum standards. You just can't know for sure, regardless of how many references you compare, without an external validation or your own quantum standard.

[Rax]

The 932B, is that a newer, unreleased version of the Fluke 931B RMS differential voltmeter?

Thank you very much for your thoughtful points. I don't know, though, what to make of your assumption the typo I had in my post was more likely of the third digit in the model number, than the much more obvious first? In case there was any doubt about that, I was referring to the 732B, which is in the business of sub-ppm voltage standard duty. I don't see how a 932B (which doesn't exist) would arrive in this conversation at all.

[Rax]

Also differently put, the fact that a reference (FX) adjusted decades after these two above agrees with them as closely as it does (my own measurements of the FX) points to the same. The likelihood of all these references across years of adjustments and checks to be wrong by the same infinitesimally small amount diminishes the likelihood they're all wrong (a statistical formalism statement) by an amount proportional to the size of that differential (smaller differential = higher likelihood of accuracy). Therefore, it's likely that they're right (I also posit the two are logically exclusive).

While it may be plausible, you just can't make any definitive statements based on that. Certainly not at the sub-ppm level. You're operating at the edge of what's physically possible with zener references. If you had a series of comparisons against a Josephson voltage standard (see Fluke article on predicting drift of voltage references, which are much easier to predict for reasons I just mentioend), then it would be a different matter. But you really can't claim sub-ppm uncertainties based on some hand-waving.

A couple of further notes on that.

• In your quote of me above, if you read carefully, you see there's no sentence that includes definitive language, so I think we agree on definitive statements, but apparently not on close reading of each other.
• I do have a few measurements against the FX, which I believe was measured against a JJ standard, and which I'm using for substantiating my thoughts.

You're making some really insightful, well documented points - thank you very much for that. But overall, really, all of my notes - observations and commentary around them - should be taken as just that: observations with commentary. I don't claim to have a JJ array tucked in a corner in my garage. But based upon my observations and a few other voltunuts from this community and other communities (not unlike I am aware of other such communities in different regions or countries, and their own "relative volts" and whatnot, making similar observations, relative comparisons in their groups with their gear, and oftentimes "sub-ppm" comments), their calibrated instrumentation, TiN's reference, and my local cal lab, I have a fairly reassuring evaluation of my bench volt, and I'm fairly certain I have exceeded what my local calibration lab can do (DCV-wise). I am "fairly certain," just about how much they're off too.

Sub-ppm graphs and evaluations are pretty common-place on this community (including my own data contributions on other threads). Some relative, some not so much.