Just a quick update - I did finally get some info out of Fluke that I can share here. Disclaimer: At sub-ppm uncertainty you will have to get your arrangements setup directly with Fluke - this is not an exhaustive description for every lab situation.
Mind you I had to pester Fluke a few times to get real information on real uncertainties, and my questions had to get "Escalated" a few times through the tech support system before I could talk to someone in the lab at Fluke Park, so this is the new corporate Fluke policy - it's a little harder to get real information compared to a few years ago. I still don't have all questions answered, but I'll wait for another time.
Also my research into this has nothing to do with accreditation, ISO17025 / NVLAP etc. I was on the search for the actual processes involved.
All the info provided before is accurate on the uncertainties you'll get on a 732 a/b Cal certificate. As "standard procedure" you'll get 0.3uV/V uncertainty on the lower cals, and 0.2uV/V uncertainty on the primary lab cal. Everett Service Center group does the lower level cals and repairs, Fluke Park Lab down the street does primary lab level cals.
IN ADDITION - IF your unit has gone in for several other calibrations before (like for at least three consecutive years prior, might be longer) AND it is deemed stable enough, your cal certificate will show a lower uncertainty of measure, not lower than 0.06uV/V, and will not be lower than the 95% uncertainty of measure while your unit was at Fluke. This is your snapshot absolute value and uncertainty for your long term drift rate calcs as compared to Fluke's reference. The uncertainty listed on the cert will be where you -start- adding all the other documented uncertainty historical rates for this unit over the years while it travels back and forth from one location to another. For true traceable uncertainty on a single 732 unit as shipped to your lab ready for use this uncertainty will probably be around 0.3uV/V (plus or minus, depends on your situation and drift/stability history) and when you are ready to start doing your first transfers / comparisons. This can be lowered over time with more cal cycles and more Vref's to verify.
So yes, CalMachine and dacman have very good, predictable 732's. That's the only way you get .06uV/V uncertainty on the cal certificate. They have had successful repeat trips to Fluke calibration and they have developed a good documented history of performance.
The 732 is a good predictable reference for sure, but that description applies to around 95% of 732's built. That leaves about 5% or about 1 unit in 20 (about) that might not be as predictable as the rest, and that's why the real way to validate Vref performance is to keep measuring against a JVS over time: you always want see how close the Vref predicted value measured up to the real calibration measure.
On the question on why CalMachine's Cert is showing that it was measured against four 732a's and not the JVS - that was an easy answer: Those four 732a's are the "Fluke Four" - these are some older very well documented 732a's that are legendary around the lab as having extraordinary performance and their predictability calcs are spot-on when they are frequently calibrated to their JVS. In other words they are "stellar performers".
This way they can use the Fluke Four 732a's working on keeping the production rate high on 732 primary calibrations when the JVS is undergoing its own maintenance and diagnostic procedures.
Fluke also offers a service on well known & documented 732's where they will print on your unit's cal cert the expected absolute value + uncertainty on the 1st of each month over the next year, but that is not accredited.
By the way - Fluke's JVS uncertainty is currently about 10nV/V - what they list on their online scope description is slightly out of date.
This person also talked about a new 732C model in the pipeline - the hope is that they can provide lower than 0.06uV/V uncertainty measures on that design - but none of that will happen until it is proven over time.
I also asked about the real usable uncertainty available with their DVMP - and they prefer that to be saved for a conversation directly with Fluke, because every lab will have a different situation (number of references, drift history on those references, etc.). If you're unfamiliar, the DVMP process involves Fluke sending you a 732b to your accredited lab, you pull some measures from Fluke's 732b and then they look at the test results to see how you did. If you pass you can join the DVMP club.
From then on Fluke sends you a 732b that you pull your transfer measures to compare each of your references, then you ship the traveling 732b to the next lab in the group, round-robin style. When the 732b gets back to Fluke it is re-measured, and then Fluke tells each lab what the absolute value is of each of the references measured.
As I said before achieving true traceable fractional ppm uncertainty at your lab location is not trivial, nor is it cheap. It is do-able but the lower uncertainty you go the required calibrations and documentation labor cost goes up. At some point it becomes more profitable for a lab to just purchase and operate a JVS of their own, and maybe sell JVS measures to other nearby labs to help offset costs.
Every situation and requirement is different - use what works best.