What is convincing is the absolute power, which is 250 W with the R20, and 108 W with the R19. Why do you find 250 W convincing but not 108 W? Apparently you have set some boundary here, somewhere below 250 W and above 108 W, where the reaction becomes unconvincing. Where is the boundary? Why is it unconvincing? To put it another way, the 250 W reaction produced a temperature difference 11°C warmer than the calibration. The 108 W reaction was 5°C warmer than the calibration. Why do you find 11° convincing, but not 5°C? Do you seriously think there is more likely to be error measuring 5° than 11°? These instruments can measure 0.1°C with confidence.
I guess this explains why we have different views of calorimetry. 250W is X6, measured relative to 50W input. R19 is X1.5, measured relative to 100W input. The key error sources are all geometric (fractions of total power) not additive (fixed power errors). Therefore we have 500% vs 50% excess in these two cases to compare with error bounds. If you don't understand that in this case it is ratio of power out / power in that must be compared with error bounds, not power itself, you will misunderstand what matters. There are also additive errors, but they are much smaller than the multiplicative ones, so much so that they can be ignored.
What is convincing relates to the way in which errors can be bounded for given results.
In Mizuno's case, the error is nothing to do with the resolution (or accuracy) of the temperature TC, which is generally more accurate than the other things. There is a possible error which is heat conduction to the output TC from an output bracket hotter than the cooling airflow. This is difficult to evaluate but obviously if it were an issue it would be much more than 0.1C. There is no information in the paper that allows me to check this.
The key absolute error sources are:
- input power measurement errors (if measured as you have said in some cases on input side of PSU). A shame because the exact input power is easy to measure accurately, and I suspect in many cases was measured accurately. Note that this cannot directly increase the absolute power measurement unless the power reported is adjusted with the PSU efficiency. This might be done, we would not know. If it was done the error bound needs to bound changes in efficiency with different conditions. However, positive input power measurement errors during heat loss calibration can cause heat losses to be underestimated.
- airflow errors
- variation in heat loss with changing conditions - e.g. reator size and placement (this need not be considered if the results used are absolute and ignore calorimeter heat losses. I'm not sure whether Table 1 in the second paper showing R19 excess power adjusts for estimated heat losses (paper method 3) or not (paper method 2). If it adjusts then the absolute R19 excess is only about 25% max. The paper implies these are adjusted figures, because the reactor heat loss calibration is shown before the results, but maybe not?
Obviously if we could bound each of these errors, we could multiply them to get overall bounds. My confidence would then relate to how close the lower bound was to no excess heat. I'd want at least 10% clear to account for errors in these bounds. I'd want another 20% clear to account for known sensitivity to room temperature change (unless this is also measured and bounded). My approach is to be very conservative unless I have enough information to make precise bounds.
X1.5 looks not entirely safe. (R19). A pity, because it definitely could be safe.
X2 looks safe
X6 looks very very safe (R20)
We could do much better with much more info about the experiment, how it was instrumented, from where the various calibrations come, because we can't be sure the given results use the exact same equipment as was calibrated. With checking of and care I'd expect 10% accuracy to be achievable, but we do not have that. Without the work being done by the experimenter error bounds tend to be a bit hand-wavy.
Rigorous record keeping and methodology - e.g. every recorded result specifies precise instrumentation and setup used to obtain it, and references specific calibration data similarly specified - would make me more confident, given the things we now know which indicate a conflation of raw and calculated data.