Please, Over and over again: Do as what LTZ datasheet tells you: '1013 amp in DIP package is right in the sweet spot for low current noise and high stability. NO available amp is better right now for LTZ. Don't use an AZ amp for best long term low drift. Follow the lead on the most successful 8.5 digit DMM - the 3458a. Do you notice the -02 option High Stability Vref board is no different than the standard issue board?? It's just a selected LTZ chip, because that is what dictates the drift, mostly - and far beyond your control. Even the expensive high stability board uses the same plastic package amp as they have for 10's of thousands of instruments for decades...if there were a terrible problem with that, we'd know. The Voffset of the amp has no effect on output voltage as it is used as a current driver. It is not uncommon for 3458a's to have drift on the order of a ppm per year and they are decades old, and that tells you how the circuit works. We have a few of those 3458a's in the lab, and all they do is work for a living, 24/7. The weakest part of that instrument is the display - if they had built it with LED's, a lot of those would never need parts replacements.
All the elements that need a hermetic seal is already inside the LTZ can for you. As long as the resistors are fairly stable, the they don't need to be high tolerance, or oil filled or expensive. Stay away from SMT for precision circuits (in general), especially ceramic caps After building hundreds, I can tell you: no real need for a fancy board, no need for crop circles, no need for voodoo slots, no magical guards, no sockets or anything like that unless you have some specific thermal flow requirement. Keep it compact and practical, use star grounds. Plan for a cover on the LTZ chip on top and bottom to keep air drafts away. If you over-insulate the thing it won't be able to regulate it's own heat. Build it and start letting it burn in for several months before your even start measuring it. Don't touch it with a a soldering iron after that. Just move on with whatever is downstream of the LTZ.
If you're going to build one board it's no more effort to built 5 or 10, and if your time is worth anything the cost is nothing compared to anything else you're doing. Select the best performers for yourself and give your buddies the rest. Real precision and low ppm's do not come cheap. The only "Poor Man" about this level of precision is what you'll be after you start doing it. Guaranteed.
Most people here won't even have the resources to measure down past 2ppm absolute anyway, and you'll never really know how good your LTZ circuit is until you have something really better to compare it to - which would be something like at least three cal'd 732's and/or four ~ five cal'd 3458a's, etc. In fact, I'd say that if you don't have those resources, I wouldn't bother building an LTZ, but do whatever you want. You need to realize that it is -very- difficult to measure a finished LTZ circuit to even 2ppm absolute accuracy, if you're doing it right. If you're going for 1ppm absolute accurate, then you probably have local access to a JJ-Array, or a lot of 732's
I'm sorry, but my opinion is: if you need a GOOD precision Vref, quit wasting time "digging the ore" and build a known good circuit for far less effort and cost - and put that energy into the other much more difficult parts of your application. You're not going to build anything better than a basic LTZ if that's what you need. Once you start looking at other areas of precision circuitry, there are MUCH more difficult areas to look at - like converting the stable 7V into another voltage for instance. That's not trivial. Dr. Frank has one approach here but there are other ways also. Or building a front end for that high resolution ADC that matches the stability of the LTZ, etc.