Thought I'd share my build of the simple LM399 based transfer. It uses the self-biasing circuit posted in the first few posts.
I used a double-walled soup container/flask from Ikea as a housing and SMA connectors for unbuffered 7V zener voltage and 10V main output.
The pink stuff is packaging material from Farnell that I used to isolate the board from the metal walls. At the top of the board sits an LM317 and underneath the white foam is the LM399. The gain setting resistors are all located below.
The blue coated resistors are Vishay PTF56 (5ppm/K) that are mounted on little standoffs. That way I didn't have to bend the legs. Unfortunately, I had a bit of brain-freeze when ordering them, so I needed to parallel them up to get the right values
The back of the board only holds the LTC2057 and some more foam.
Currently the unit sits in a lab corner powered 24/7 to observe the thermal behavior. Its intended purpose is momentary measurements so it's not going to be powered for any extended time. But I thought it'd be interesting to see whether the LM317 generates enough heat to upset the LM399 or influence the gain setting resistors. After half a week I opened the lid and the inside was a few K warmer than the outside of the case but not by a lot. For the next build I'll try some different things.