Your description of how it works is not quite accurate. This device does use things such as an estimated drag coefficient and rolling resistance (both of which it gets from the calibration ride) to do its calculations. So changes in position and/or equipment do affect the accuracy of the measurement. The conditions will have to be somewhat extreme for the error to be 10-20% though, especially if you look at longer time intervals. During instantaneous measurements it could be off by such amounts, but the same occurs within different 'real' power meters.

One neat side effect of the above is that if you have both a 'real' power meter and a power pod, you can do aero experiments on the fly to determine how many watts changing position saves you or what difference those $2k wheels really made.