Axil's comment seems a little off the wall, but the paper he refers to by Cravens is pretty good. You should read it. See:
Hi Jed. Thanks for this. I read an earlier account - this has a little more information.
The thermistor datasheet shows 2K at 25C . With 250mV that is 100uA current. Or for 250ohm (80C) 1mA. Much higher than the recommended current of 10uA. However the main problem here is self-heating of the thermistor relative to air, with a time constant of 10s and thermal resistance of 1C/mW (much more significant than the claimed ball to bath resistance of 4C/W). The test noted in the write-up does not determine this, but at 250 ohm and 250mV we have only 0.25mW - this error possibility can be dismissed if the given data is correct.
With the active thermistor in a D2 atmosphere the obvious issue is permanent drift due to chemical contamination. It is encased in epoxy - but that would not be enough to prevent D2 diffusion. This fits all of the data and the standard check of temperature stability after the test has never been done, as far as I can see. Also, metal oxide thermistors would likely be affected by exposure to hydrogen (thermistors without coating are also used as deoxygenating gas sensors!) From the datasheet:
Stability and Drift
While thermistors are generally very accurate and
stable devices, conditions such as over-temperature
exposure, humidity, mechanical damage or corrosion
can cause uncontrolled changes in the resistance vs.
temperature characteristics of the device. Once this
characteristic has been altered, it cannot be
re-established.
I think this is a pretty good high likelihood explanation for these results. Of course, were this a serious test, a before and after sensor calibration would be done, and this detected.
From the write-up:
It was interesting that only two people had serious doubts
about the heat. It is just hard to argue with a 4 degree signal
that is hotter than its surroundings.
The only way to determine this 4C difference was the thermistors. They were epoxy bead - not designed for industrial environments, and not checked after the test (at least it does not say this).
4C can easily be detected by touch. But not up at 80C where only transient touch is possible and so experienced heat varies with touch time and thermal resistance.