Hm. So then if HWt shows 86W and I'm spinning with 20 RPM (it does), then torque is 86W / 20RPM = 4.3Nm. Then for PAS, the CA calculates 4.3Nm * 50RPM = 215W. Then that's fed in the other formula, (215W - 85W) * 3 = 390W. If this calculation and logic is correct, then that could explain the power spikes above the start threshold with the Sempu. This does match what I'm observing - significant power output when spinning slowly and with only the weight of my legs. This seems to be entirely driven by the bogus force output of the Sempu generated by leg weight instead of spinning torque (which it can't detect directly). Seems like so long as the scale factor is high, this effect is inevitable. One way to counteract it I came up with is setting the starting threshold to 1W and then using smaller scale factors. That way the bogus leg weight torque isn't multiplied as much and the response feels less disconnected from my legs. I guess another thing to try would be to set the zero force level for the PAS device a little higher.
Is there a way to adjust the 50RPM value used to calculate PAS human watts under 50RPM?
If you apply 86W of power, your assist with an 85W Start setting and 2x scale factor will be (86HW-85HW)*2W/HW = 2W. As you can see, the start power is affected by the scaling factor. So, at 2x scaling you need 170HW to overcome the Start level.
Sorry I didn't understand what the last sentence means. Is start level the start threshold (85W)? If so how does scale factor affect that?