Certainly not the crank length - that has no effect whatsoever. As the pedal motor is applying a pure torque to the crank via the pedal axle it could be doing so anywhere along the crank and the effect would be the same.
I was assuming that the pedal would be/is quite a bit more bulky than a normal pedal, with the battery packs each side. So the overall pedal length front to back could be about 6"/15 cm. That ties in with my memory of the prototype pedal from the show. It could be a bit less.
The lever arm I used was this entire length. My reasoning was that the limiting case is when the foot just lifts off the back edge of the pedal (and hence, assuming a flat rigid shoe sole, throttles back the motor via the pressure sensor close to the pedal axle).
At this point IF you neglect the weight of the leg (which you can't - on an unpowered bike there is always weight on the pedals - it balances out though because the two cranks are at 180 degrees) then there would be 10 kg of force between front edge and your shoe, and 0 at the back edge. As long as your leg weight and pedalling force provide over 10 kg the pedal stays flat.
Say your leg weight and a bit of pedal force provide 30 kg in total, applied only on the front and rear edges. At max assist there would be 20 kg on the front edge and 10 kg on the back edge of the pedal. So a difference of 10 kg over a distance of 6", i.e. enough to provide the necessary reaction torque.
I can see why using half of the pedal length is attractive, but it's the slightly counter-intuitive fact that you can apply a torque anywhere on a rigid body and it has the same effect which applies here I think.
If in our assumption the two contact points between pedal and shoe are the front and back edges only, then it must be this distance which is used for the torque calculation.
Even if I am wrong there and the forces involved are double my estimate then it's still a feasible proposition IMO. At least feasible enough not to be dismissed out of hand.