PREDICT · DROP · COMPARE
Galton board
Where do the balls land?
Each row makes an independent left/right choice. Motion illustrates the route; it is not a collision simulation.
Does more data always look smoother?
Inspect a prefix of the same seeded experiment. Larger choices extend the target if needed.
From independent choices to a distribution
A ball makes n independent decisions, each with the same probability p of going right. Its bin K is the total number of right turns: 0 is all left and n is all right. This is a binomial model, with mean np and standard deviation √(np(1 − p)).
Hide the theory first and predict the distribution. Try p = 0.5, then move the probability toward either edge. More balls often make proportions more stable, but no finite sample must match theory exactly, and individual bins need not get closer every time. Not every binomial distribution is bell-shaped.
This ideal model ignores ball collisions, friction and peg geometry. A fixed pseudorandom generator makes experiments repeatable; it is not a source of physical randomness. NIST: binomial distribution