LitterBug Motor Input Output Curve Fit at the bottom of the output pane.
Real ESCs/motors are not linear; PID expects linearity; this attempts to fix up for that. We apply only a small correction (x0.9) by default, when 0.85-0.80 is probably correct in actuality for most setups.
Our philosophy to under-correct when adding new mechanisms rather than over-correct and risk introducing new degenerate behavior, so the default of 0.9 is not enough correction. Basically, it raises the motor desired-power command to the 0.9th power before converting it to an output range and sending it to the ESC.
Hovering over the control offers some guidance, but----- there tends to be more of a difference in power when an ESC is commanded from 80->85% power than when, say, it is commanded from 20->25% power. So the same correction for a given error may be overpowered and oscillate at higher power settings, and undertuned at low power settings. (This used to be addressed crudely with "Throttle PID attenuation" where control coefficients were turned down in response to high throttle).
Nonlinearity can also cause cross-axis coupling-- roll to pitch, or pitch to yaw, etc. Of course, once the control system "hears" this it can apply a correction, but it's better to minimize these kinds of errors int he first place.