Bob cast his SGE in the form of a wheel. This introduces the feature of rotation which is a "red herring" for some people. They think that the paradox somehow depends on rotation or requires consideration of centrifugal effects. It doesn't, as this non-rotating bouncing ball engine illustrates.
A ball bounces up and down between floor and ceiling, both rigid and massive. The bounces are assumed elastic, that is the ball's velocity after impact is the same as before impact, but with reversed direction.
Now imagine that the gravitational constant g is slowly but steadily decreasing. The ball is released at rest from the ceiling. The ball attains a certain speed when it reaches the floor, and rebounds with that same speed. But since g is now smaller, the ball still has a small velocity when it hits the ceiling. Clearly this means that on completion of this ceiling-to-floor-to-ceiling cycle it has gained a small amount of kinetic energy, which we could extract with a slightly inelastic ceiling panel. The panel would steal just that extra amount of energy, bringing the ball to rest there momentarily. The ball would then start the next cycle with zero speed, as in the previous cycle. The gravitational force, though slightly smaller than before, would cause the ball to fall to the floor and bounce back to the ceiling, where we again steal the excess energy gained in this cycle, and so on forever, or until gravity runs out, whichever comes first.
The assumptions of perfectly elastic impact and infinite mass floor are no more unreasonable in posing this apparent paradox than the assumption of frictionless bearings in the wheel. Given these assumptions we still ought to be able to analyze the machine and show whether it could work as claimed.