I am still testing against my Elite Kura and Garmin Vector 2s. I calibrated the scale with only one known weight and assumed a linear relationship between torque and weight. It is close, but not close enough. There is a significant deviation between riding standing or sitting at the same power.
The plot below shows the torque profile for one crank revolution. Standing and sitting cadence and power output was identical (as measured with Elite Kura and Garmin Vector 2s). However, my power meter measured 230W for sitting and only 220W when standing.
The integrals for red and blue curves should have been equal, but they are not. The in saddle position gives a more smooth torque profile with a peak around sample 25. The out of saddle position peaks at sample 35 and differs in shape. The blue curves are only slightly higher than the red ones, in order to give the same power reading they should have been giving higher amplitudes at the peak. I assume that for higher torques the crank bends less and less.
So, we have some nonlinearity here. I already included a second order term in my torque calculation which takes care of this.