Hi Itz2000,
This is a little speculation on my part plus some theory concerning the innards of a SR-104. While it is possible that they are using four resistors in series/parallel configuration, I think that may be more trouble than it is worth to build and match. Theoretically, since Vishay has used it in at least one of their wiz-bang resistor types, it is possible to use two resistors in series (in this case, two 5K resistors) with matched opposing TCRs to 'cancel' out to near zero. Out of a given batch, say a number larger than 50 resistors, there is a fair probability of finding a reasonable number of equal but opposite TCRs to build near zero TCR sets out of. At least with Evanohm, there are no hyperbolic TCR curves to mess with.
This may or may not be how the SR-104 is built, I've never seen the insides of one either and because they are very expensive I doubt that anyone outside of the production line has seen the insides of an SR-104. The most recent descriptions I've read of the SR-104 resistor sounds more like there is only one 10K resistor inside there as the primary, the other 10K resistor is strictly for internal temperature measurement. It is certainly not out of the question that there is only one resistor in there instead of two.
Fortunately, since an SR-104 deals with very small temperature excursions, this makes the matching game a bit easier and if the TCRs are reasonably linear around the cardinal point, you get a very nice, near zero TCR as a result. That is the relatively easy part of it, the other part is the conditioning of the alloy/bobbin (mica in this case) to play nice with each other. While it is possible to get wire with zero (or near zero) TCR, when it is wound onto a bobbin (whatever the kind) things do change and only if the remaining processing does the right things to the wire/bobbin assembly does the resistors turn out with the desired end TCR. The variables involved are many and most of them cannot be controlled real tightly so there are always variations in the end product no matter how good the total processing is. Just about every process done to a resistor, right up until the final calibration, imparts stress or removes stress, the end product, hopefully, has had most, if not all, of the stresses removed or compensated for.
Yes, a metallurgist told me that alloys can generally be recycled as long as they are not contaminated so wire that misses the TCR mark can be cooled down and put back through the heat treatment process again to try and hit the mark. The tighter the 'mark', the harder it is to hit.
That impression about the mica card making it possible to withstand large temperature variations was more smoke and mirrors, perhaps in their minds mica did that but I don't use mica and my resistors withstand even wider temperature swings which would likely wreck havoc on even an Evanohm standard, but yes, an Evanohm standard will certainly take more abuse than a Manganin standard with much less damage to it.