My starting point: I work in a resistance lab. I have measured many old commercial (Tinsley, Guildine, MI, Fluke, ZIP, Siemens&Halske, ESI, Leeds&Northrup, General Radio, HP, Cambridge, Wolff, Hartmann&Braun, AOIP, Cropico, Sefelec, IMMS, Tettex, Göerz, Quadtech, Willow...) standard resistors. Some of them are very good even today. On the other hand some have TCs of several ppm/C. Even they are mostly good if kept in oil bath. Their drift is typically in the range of 1 ppm/year. There are much worse ones too, of course.
In the sets of our own reference standards we have some values, for example 100k, that we actually have only one really good reference resistor. Some day it may break down and so we need some backup. It would be good to have one or two sets of secondary standards that I keep all the time in our +/- 0,1 C air bath. Then one or two more sets that I can give to my colleagues when they need a reference resistor. They may not have so good temperature control in their setups, maybe 23 +/- 2 C.
As yoy know, nowadays there are resistors with amazing stability and TC specs. Why not to by a good Vishay resistor with $50 and put it in the box? An alternative is to buy several commercial standard resistors, maybe $4000 each? Not very attractive.
I have already put some Vishay resistors in a box in 2012, but now I have a new 2018 design which I will now describe.
* Enclosure
Die cast aluminum box. Connector for grounding/shielding.
* Component selection
From 100 ohm to 100 kohm I think that the best ones availlable are Vishay VHP100T (100 and 1000 ohm) and VHP101T (10k and 100k) oil filled hermetically sealed 0.005% tolerance resistors. I'm also interested in other decadic values between 1 ohm and 10 Gohm and there might be some use for up to 100 Tohm and down to 0,1 mohm, but they are so different worlds that they go off this topic.
Data sheet:
http://www.vishaypg.com/docs/63003/vhp100.pdf* Trimming value
Vishay VHP100T and VHP101T I have are mostly +/- 50 ppm from nominal value, so I see no need for trimming. There's no difference if the deviation is 50 ppm or 0,5 ppm, as far as it's known. Also, even if the initial value would be trimmed to for example less than 1 ppm from nominal, it would drift out from that window over the years. Or maybe not? VHP resistors should be very stable also in long term.
* Compensating TC
If the 0,05 ppm/C specified for VHP-resistors is true, there is no need for TC compensation. In a temp controlled calibration lab the room temperature variation would be 1 C, in air bath maybe 0,1 C and in good oil bath 0,01 C.
* Attaching the component and selection of external terminals
Common path for F(orce) (current) and S(ense) (voltage, potential) leads should obviously be short. Everything between sense terminals and actual resistor will be measured as a part of the resistor. So the layout for the connection is F1----S1---R---S2----F2. Sense terminals close to the resistor. So in practice there are 4 binding posts linearly.
Sense terminals are gold plated, Force terminals are regular not-gold-coloured-material, I didn't check what was that.
In practice, does the length of the leads from resistor to sense terminals matter? Some quick calculation: 2 cm of 0.635 diameter copper wire has a resistance of close to 1 mohm. So it's 10 ppm of 100 ohm. More important is the TC of the copper wire which is +4000 ppm/C so the value of my 1 mohm copper wire changes +4 uohm/C. Combined with 100 ohm resistor this changes the value of 100 ohm by 0.04 ppm/C which is in the same range with the VHP specs of 0.05 ppm/C. So it matters a little bit. For resistors higher than 100 ohm it doesn't matter any more.
No soldering is used in sense connections to avoid heating the component during assembly (this was probably unnecessary, a component should be designed to be soldered) and to avoid any thermal voltages due to soldering junctions.
Instead of soldering, the sense connections are just pressed under the gold plated nuts of the sense binding posts.
Force connections are soldered.
In my older design in 2012 I (gently) presser the resistor to the cover of the aluminum box with a strip of copper which was attached the the box by screws. That time I thought that it would be good if the resistor is connected to some large metal mass to stabilize temperatures. Maybe also some heat sinking ideas were present.
Now in my 2018 design I decided that no other mechanical connection in addition to the binding posts are needed. It may be better that no mechanical stress is applied to the metal can of the resistor. No heat sinking is needed because the resistor is intended to be used in very low power. No larger thermal mass to buffer temperature fluctuations is needed because the TC of the resistor should be practically zero.
* Temperature measurement
In my older design in 2012 I put a 10k thermistor which was (gently) presser against the Vishay component. I wanted to measure the actual temperature of the resistor directly from it. In my 2018 design there are several resistors in the same box and I would have needed same amount of thermistors and 2N more binding posts for them. Also some mechanical stress, which I want to avoid, would have been inevitable if I had attached termistors to the metal cans of the Vishay resistors.
Now in my 2018 design I only have one 3 mm hole in the middle of the top cover of my aluminum box. An external 3 mm PT500 sensor is put through the hole and there is a simple mechanical guide structure inside the box that ensures that the PT500 touches the floor of the aluminum box. 3 mm PT500 is also practical selection because our resistance lab uses them in our countinuous ambient monitoring system.
* Measuring the resistance value in general
Fortunately our lab has a commercial resistance bridge made by Measurements International so repeatability of <0,1 ppm is possible and absolute uncertainty of below 0,4 ppm can be reached. Measurement power should be < 10 mW, preferably < 1 mW to avoid any self heating.
* Long term stability
Some tests how a fresh resistor behaves should be done. In the data sheet there is interesting discussion about the three steps of post manufacturing operations of Vishay: "The exercises that are employed are (1) temperature cycling (2) short time overload, and (3) accelerated load life." Operations 1 and 2 are done to all resistors by default. For number 3 it is written: "How much acceleration is a function of the application and should be worked out between our applications engineering department and your design team." So the operation 3 is not done for the components by default?
Encouraged by this I actually sent an email to Vishay and asked could I do this accelerated load life by myself. (No proper technical reply for that email so far.) What is it in practice? I made a long 17 days continuous measurement for one of my 100 ohm resistors immediately after it was taken out from it's package and put in the enclosure. It was no surprise that there was more drift in the beginning and the rate monotoniously decreased. Total drift was -0,8 ppm over the 17 days. During the last day the rate was no more than -0,024 ppm/day (-8,6 ppm year). If I fit an exponential decay curve on the data it seems that the drift would settle to practically zero after about 200 days of loading.
Total drift over that period would be almost 3 ppm, which is not in specs, but it doesn't matter if the resistor is stable after that. Maintaining resistance values is the work of tens of years, not days or months.
So I started to think should I put some current over all my resistors for 200 days to age them to their most stable part of life
Then I found some other info from here:
https://www.rhopointcomponents.com/media/blfa_files/VPG_Design_and_Selector_Guide_for_High-Precision_Resistors.pdf The text would seem to be in conflict with the data sheet, because it is written that "STO (Accelerated Load Life) is performed on all resistors during manufacturing, with a function of eliminating any hot spots if they exist." So according to this the PMO number 3 is also done to all resistors and I shouldn't see this 500 - 1000 hours of initial drift? Maybe the drift of 17 days I saw was due to mechanical stress during the assembly. That would be more convenient because I just have to wait. No need for putting some "ageing current" over all the resistors!
After this initial 17 day experiment I have been measuring other resistors of 100, 1k, 10k and 100k which were all assembled at the same time, so in other words their assemblies were 17 to 22 days old when they were measured for the first time. During 3 to 16 hour measurements I have seen no drift more than 0,05 ppm which is just noise.
Conclusion: 0,8 ppm drift over 17 days was most likely due to stress caused by assembly or just a bad individual resistor. More long measurements is anyhow needed to verify this.
* Summary of 2018 design
I ended up to a very simple, even primitive looking construction. But I hope that it's well resoned mostly because of the practically zero TC of the component. If the TC is not zero, then you have to start thinking bout larger thermal mass, maybe some double enclosure with thermal isolation between them, using thermistor attached directly to the resistor etc.
* Long term stability of my 2012 design.
In 2012 I made 4 boxes with one 100 ohm Vishay Z201 in each of them. I have given away 2 of them and don't know how they behave, but the two others have quite linear drift of -1 ppm/year and -2 ppm/year. So they are better than many commercial standard resitors which may have a price of $4000 or more!