Let's start with the questions about the G.R. resistor first, the rubber silicone filling in theory was put there to help absorb stress between the windings and the bobbin/shell, the problem I found was if the shell is overfilled with the silicone (i.e. no gap between it and the shell) it does just the opposite. After quite a bit of digging I found that silicone rubber has an COE of 900 microinches/°C, somewhat higher than I thought it would. So as the resistor gets hotter, everything is expanding and within a certain range, the silicone does absorb some of the stress but after that point it begins 'spreading' the stress around from the expansion against the outer shell and actually increases stress on the windings (even if you've cooked the resistor silly). I found through actual measurements that this stress takes quite a bit of time to relax and even over a period of days just sitting, the resistors did not return to the original starting points, albeit the differences became smaller with time, almost none of the test resistors went back to the starting resistance. I also had these tests confirmed by another independent tester. So that becomes another variable in the stability equation.
Next, the question of how snug the windings are, as indicated in the X-ray, if you look closely at the X-ray, particularly the lower winding, you will see rater uneven, even sloppy turns of wire, not the even turns one would expect, that indicates that the tension of the wire was essentially uncontrolled during winding and during calibration the turns in the upper pi was likely loosened by the calibrator when searching for the correct resistance point. I noticed that the wire size used in this resistor was unusually small for a 120R0 resistor, not the best idea. To some degree, loose turns are good except when they are too loose and sloppy, this can cause problems (not so much in low ohms), the wire must be wound with a certain minimum tension, without that the resistors become inconsistent, this tension is taken care of later on. In the resistor X-rayed, the less than great winding may be because of an inexperienced winder.
Also a problem is kinks or loops in the wire (often caused by insufficient tension while winding or by jerky feed of the wire from the spool), there appears to be a rather sharp bend in the wire at the top weld joint, these can affect the TCR, another point is that G.R. uses alloy 180 ribbon to connect the resistance wire to the lead, this is used by multiple manufacturers as an intermediary since Evanohm cannot be directly welded to a copper lead. While it is a relatively small resistance, it is also of a higher TCR than Evanohm and is in series with the resistor at both ends, this will cause a slight hyperbolic curve in the TCR line particularly with lower resistances. The main reason for using the ribbon is the attachment between the lead and Evanohm, as far as 'relieving' stress on the weld, yes and no, while molding the lead/ribbon into the bobbin does substantially prevent external applied stress on the lead from getting to the weld joint, it doesn't protect the weld joint from stress applied inside the resistor and like all welds of this tiny physical size, various factors come into play that can affect the weld joint, in turn affecting the overall apparent TCR of the resistor.
Not to throw bricks at the competition, but I'm trying to illustrate the complexity of the interaction of all the materials that can go into the making of a precision wire wound resistor and also applies to the film/foil resistors as well, there are no absolute solutions to completely fix any of the various causes of stress on the resistor/element. Each set of materials has its own characteristics and interactions and when you're working in the lower PPM range, everything becomes important and more difficult to adjust for.
Finally, variables change with wire size used, larger wire sizes must be handled differently than smaller wire and the techniques used for winding varies with the size, how the wire is wound onto the bobbin can also affect the resistance. One other 'myth' I'll pop while I'm at it, the misbelief that reverse winding the pi has a significant effect on overall inductance, it doesn't, measurements indicate that while there is a small reduction in overall inductance, it is not really that significant when it comes to AC signals. The coupling between pi is quite loose and therefore has little cancelling effect because they are next to each other in series so the flux lines do not cancel significantly, not to mention that the Q of resistors is exceedingly low.