The famous experimental binomial distribution
turns into a normal distribution with sufficient rows (experiments: n).
If your pc can handle it you can make copies and show 3 machines at once.
Be warned, it is pretty slow on a fast pc if you do that.
Pretty good approach, but I think it's still biased by 2 things: momentum and sim frequency
If you made each of those gears into circles and made a script that made each ball get vel [0,0] and pos right upon the circle (To align the balls so that Algodoo actually does this randomly) and cranking the simulation frequency up to 120 or 200 Hz, there might be different results here
That wouldn't be very easy to do, but I might try something, i dunno
