There is a strip of duct-tape covering the metal edge between the insulation and the machined surface. Although it is more difficult to model (what is the thermal conductivity of duct-tape?), measurement shows that it reduces the heat flow from the non-test surfaces.
I finished the interrupt-driven code to measure the rotation rate of the fan. Each second it divides the number of times a blade has passed the sensor by the length of time between the zeroth and last blade passing. This reduces quantization error such that most readings are stable to within a couple of r/min. Occasionally there is an extra interrupt causing one reading to spike. It may be possible to correct the spike in software.
Measurements with the impeller anemometer are used correlate the fan rotation rate with the wind speed inside the tunnel. But the two significant digits of the anemometer are limiting. We tried to interpolate the dithering of the low order digit in collecting the data shown in this graph.