But how can we arrive at a predictive model from such measurements? Even if you measure "all" ambient conditions like temperature, humidity, air pressure you may have delayed contributions (typical with heat conduction), integrators, differentiators and nonlinear effects and so on. Did you ever try to fit a model that (roughly) predicts the behaviour of such resistors? I think in order to do that one should define certain key tests to isolate parameters like the TC curve, one by one.
(Slightly late reply...)
I wasn't thinking of that sort of simulation at all. It's just that in the limited tests that I've done, the effect of simply running at a particular temperature for an extended period can be a lot more important than the TC. At least separating out temperature and humidity would be useful for some resistors.
It's worse for new components. Page 9 of the design guide at:
http://www.texascomponents.com/pdf/TXCCBMFDesignGuide.pdfgives some example plots. The 500+ hour settling times are not too dissimilar to some other components like LS8 versions of an LT1027 and LCT6655. The reference VH102ZT in the oven at about 45°C took around a month to get to the point where I don't see it drifting down in resistance, even although the oven was powered for months over winter, and the room was mostly 24°C+ during the summer. The plots in the design guide don't show convergence, which seems to fit with both the PMO info that I've seen mentioned and another extended settling time if the temperature increases again, but I've not checked the extent to which that happens.
I could probably reduce the temperature a bit to be more realistic, and I really need to test low temperatures too. Running extended tests at different temperatures would help to show how much the cold-to-hot changes reverse after the temperature is reduced again for these resistors, and what temperature changes are needed for lasting effects.
For tools that got magnetized we have a demagnetizer in our lab, which works by imposing magnetic field cycles of decreasing size. Maybe one should try that to force the resistor into a state where it no longer drifts. I think this is the idea behind that Pickering patent.
For an LTZ1000, that's a particular solution for a particular hysteresis problem. I would expect that there are some hysteresis issues with resistors which are roughly similar and do respond to temperature cycling, but there are a lot of other things going on too.
Even in the LTZ1000 case it's just an attempt to get rid of an issue caused by a temperature change, not some way to improve the performance.
I noticed that Dr. Frank mentioned using temperature cycling for foil resistors that had been given large temperature changes, which does look a generally similar case, but I didn't see any details.
As far as I understand it, the 'Post Manufacturing Operations' on resistors are designed to replicate extended use in particular conditions, but they're more complex. There are various comments on that for different resistor types earlier in this thread, and page 15 of the design guide above also has some info. I'd like to get some more info on PMOs, but I've not seen many details.