I recommend reading the book

*Dynamics of Gambling: Origins of Randomness in Mechanical Systems*

if you want a good understanding of exactly how randomness arises in coin flipping, dice throwing, and the roulette wheel. In particular, Chapter 5, entitled *Nature of Randomness in Mechanical Systems*

The authors model mechanical systems in high detail to allow them to calculate the bounds on initial condition uncertainties needed to allow correct prediction. They do not model a lotto ball picking machine.

Could you model a lotto ball picking machine in sufficient detail in computer simulations on modern computers to reveal the initial condition bound uncertainties guaranteeing accurate prediction? Yes, you could.

The importance of ball collisions creating piecewise smooth systems is key to understanding how pseudorandomness arises in lotto ball selection machines. Without the collisions, the system could be well modeled smoothly, vastly reducing the pseudorandomness.

The authors claim that grazing collisions are the ones most responsible for pseudorandomness in mechanical systems. Here, grazing means the state space trajectory is nearly parallel with the non-smooth subspace of one lower dimension embedded in the state space.

If the grazing collision claim is true, then all it takes is one or two dozen such grazing collisions to get the random shuffling model to guarantee a random ordering after those grazing collisions.