Small ps: the funny thing is when the flywheel has constant speed, the Kickr knows the power you’re delivering is balanced with the power the kickr is extracting.
pps: I might well be that instead of derivative the kickr software takes a numerical integral over tau times omega with a soft of moving average, which might be less susceptible to high frequency noise in the rotational velocity. So it could be implemented in many ways, and depending on where the dominant noise contribution / uncertainty sits in the measurement system a different calculation leads to a more precise/accurate result.
Another small beside: the kickr supports cadence readout without an additional sensor. I suppose the readout from the torque is sufficiently low noise and with enough temporal resolution to detect the modulation of torque corresponding with pedal strokes. Like detecting the frequency of a small AC signal on top of a DC signal. This is consistent with the kickr not being able to detect cadence when you pedal without force…
One thing that I don’t get yet is the dependency in temperature. I suppose it is relevant as the Kickr Core upon spindown gives that readout. What bugs me too is that for me the offset always gives 32768 (with a minus? I forget), which is exactly 2^15. That suggests the value is clipping the dynamic range, which in my experience is never good in a measurement system. But ah well, I’m having fun on the Kickr and never worry about the thing, which is optimal.