I've built the first 10V prototype with the 2N4391. The reference voltage was around 7.228V, so I've used OPA227 and some wire-wound resistors to scale it up to 10V, the same type wire wound resistors were used for the reference source as well. Without much tuning I've got around 2ppm/C and very low noise level. This is the first time I've measured the JVR against the Fluke 731B, using Keithley 2015 on 100mV range. The variations are around 1ppm, the noise is well below 1ppm p-p.
This is a step towards a properly built reference - I've used a good aluminium case, good resistors and Tellurium Copper terminals, though the circuit is built on a prototyping board.
Cheers
Alex
10V JVR on 2N4391 measured for 30 min against the Fluke 731B by Keithley 2015, 100mV range, NPLC1, no filtering, 1ppm per division: