It doesn't. The belt is trying as best it can to pull the cart off the back end of the treadmill. If the cart were a brick, it would succeed. In addition, when the treadmill is tilted up, gravity is also trying to send the cart off the back end of the treadmill. If the cart were a billiard ball, it would succeed even if the belt wasn't moving. With two things both trying to send the cart off the back end, what keeps the cart from doing so, and what in fact will send the cart off the front of the treadmill if there isn't a spork there to stop it? It's the fact that the cart gets some restraint from the air, which means that the wheels have to turn because the cart isn't backing up as fast as the belt would like it to. And that turning of the wheels is leveraged into prop rotation, which drives the cart forward. There is no forward thrust coming from the wheels/belt interface.
On page 7 of the rail book it shows slip to the rear of the axle.
How can a belt do that? The 'locked' region cannot drive the wheel forward of itself, even if the were the way the wheel were driven.

