Continuation of Post 1161:
The fit is MUCH better over the Figure 4 calibration data (from Mizuno ICCF22 preprint) using the V = A * exp(Wb/w) + B regression than using the V = Vmax(1 - exp(Wb/w) regression.
I took the data from Figure 4 and ran the regression. The model using my rough data input from figure 4 came out:
B = 7.7081
A = -6.5602
w = 6.0603
vs the regression from fitting the data in post 1111 provided by Jed:
B = 7.53828
A = -6.38828
w = 5.78291
Thus, I come to the conclusion that
1) Mizuno et al used the calibration data to generate the factors A, B, and w in the spreadsheet; and
2) the data provided in post 1111 is then using the above model from blower power to derive airspeed.
I believe that the calibration is reasonable and that the airspeed values are reasonable.
I do not have a fluid dynamics physics reason why the model (V = A * exp(Wb/w) + B) works empirically, but it does work (residual standard error 0.03). It works better than a linear model (RSE=0.10), or a two factor exponential model (V=Vmax(1-exp(-Wb/b), RSE=0.15).
(A possible hypothesis is that the zero blower power airspeed B - A, could be from natural convection if the reactor is heated. An even hotter reactor would have a tendency to increase this natural convection effect so that the estimate of airspeed from blower power would be underestimated at higher temperatures, and thus the calorimeter output mass airflow heat removed would be underestimated at higher temperatures. If true, this is conservative at higher temperatures than the calibration, i.e. the reactor is making even more heat. I would prefer again to calibrate under identical conditions with identical emissivity tubes, but the error would mean the experiment is creating more heat than measured.)
Note that if Jed supplies us with the 6 decimal points of data that Mizuno used to generate the calibration (the underlying data in Figure 4), we will likely get the same exact values.