As I mention in my post, I already have a Ref01 board from voltagestandard.com. A Geller board is in order and will probably be shipped ultimo this month.
The Ref01 is a neat product and mine is within spec 0.01%. The temporary problem is that the value has drifted in trend from around 5.00022 to 5.00045 Volt in the 2 months or so I have had it. The most stable of my other references have not had this trend relative to the DMM, so this leads me to think that the chip on the board needs more aging. As you say, the recalibration is cheap and even free for two years, so it will be sent back. But there is no point in calibrating a source that has trend drift.
Also, it might be lost in the post. Or it might fall down 5 feet while handled at an airport so the trim pot changes a little bit. So I will probably remove the trim from the board before I send it back for measurement. I would rather have a number that is off "5. 000 0...." but reliable, than a fragile pretty value that can be lost more easily.
Detecting the systematic drift in one source by comparing it to several sources and also the danger of losing a particular reference in post, by accident or for any other reason illustrates imo why it is useful to have at least 5-10 refs. Having even more and spreading them out on 1, 2.5, 5 and 10 Volt also gives you a better idea of your meter's linearity. You can also combine them to get several reference voltages without resistance dividers.
To replace real calibration I will do as you say: Send my "best" references one by one to "voltagestandard.com" and "gellerlabs.com" and others. But before I do that, I need to "become familiar" with the refs. And they have to be stable. So I am in a process of may be a year where I buy a few different chips every month just to get a working idea of where 10 Volt is (I now have several refs that are better than 100 ppm with almost certainty). When I know by inter-comparison which of my 10-15 refs that are most stable, I will send them by post from Thailand to Europe or America for "postal order" calibration wherever I think a reasonable offer is given. I will part with one and one only for a given time period (change only one factor of a familiar system), and I will use several measurement services (it obviously increases the value of a measurement if two different persons / calibrated meters pins it at the same or close similar value).
I am also making a slow relay-based multiplexed LT2400 ADC to computer 20 Input Data Logger & Plotter that saves as much data you want to disk, and do all kinds of real-time and postponed statistics in Processing and C+ (using the same stuff as in the video in my post over). It does not need a more accurate ADC-voltage reference because it will only compare the noise and drift of my voltage sources. So I use whichever reference I have that is most stable as reference for the LT2400.
In time I will build better references. I have bought a single LM399 and LTZ1000 as toys to play with. Cheapbay fake or real - I don't care. For now they are toys. I need a fleet of commercial ready-to-use refs before I start to send high precision home brew voltage sources over the Atlantic or the Pacific for calibration to 5ppm or better.
So my next order is 4 x 2.5 Volt. According to spec I will get:
Low Noise: 0.25ppmP-P (0.1Hz to 10Hz) 625nVP-P for the LTC6655-2.5
Low Drift: 2ppm/°C Max
High Accuracy: ±0.025% Max
No Humidity Sensitivity (LS8 Package)
Thermal Hysteresis (LS8): 30ppm (–40°C to 85°C)
Long-Term Drift (LS8): 20ppm/?kHr
100% Tested at –40°C, 25°C and 125°C
http://www.linear.com/product/LTC6655By putting 4 of them on top of each other on continuous individual independent battery power (9 Volt rechargeable, 2-3 in parallel, change one at the time so power is not interrupted) and using cheap 0.1% 5 Volt refs as preregulators, I will get a fairly good 10 V. If it turns out to be very stable, I can modify it for trimming by using 5 chips (giving 12.5 Volt) and a divider.
The real nerds in this game baby-sits references for years. A slow tedious work with diminishing returns as you get closer to the practical boundaries for these kind of measurements ...