Can physics predict a lottery ball machine?

A lottery machine is not a computer program. It's balls, air, and bouncing — physics. And physics is predictable, isn't it? Before dismissing the question, it deserves respect: casino roulette has genuinely been beaten with physics, more than once, and the receipts are in the literature. The interesting part is why the same methods die completely at the door of a lottery studio.

The roulette precedents are real

Thorp and Shannon, 1961. Edward Thorp — later famous for card counting — teamed up with Claude Shannon, the father of information theory, to build what Thorp documents in his paper "The Invention of the First Wearable Computer" as exactly that: a cigarette-pack-sized computer, operated by toe switches, that timed the roulette ball and rotor and predicted the ball's landing octant. Tested in Shannon's basement lab, the system showed an expected gain of +44% on the most favoured octant, and the pair took it to Las Vegas in the summer of 1961, keeping the machine secret until 1966.

The Eudaemons, late 1970s. A group of physics students including Doyne Farmer and Norman Packard rebuilt the idea with shoe-mounted computers and took it to Nevada casinos — the project chronicled in Thomas Bass's book The Eudaemonic Pie and summarised in MIT's wearable-computing history.

Small and Tse, 2012. The definitive peer-reviewed treatment: "Predicting the outcome of roulette" by Michael Small and Chi Kong Tse, published in Chaos (open preprint). Using just the timing of ball and wheel rotations on a casino-grade European wheel, they demonstrated an expected return of at least +18% — against the −2.7% a random bet yields. With a camera above the wheel, prediction improved further.

So "you can't predict physical gambling devices" is simply false as a blanket claim. Why does roulette fall?

Why roulette is beatable

Three features make roulette a physicist's dream target:

1. One projectile, simple dynamics. A single ball rolling on a smooth rim, decelerating under friction, then dropping onto a slowly turning rotor. For most of its flight the motion is nearly deterministic — measure position and velocity, integrate forward.
2. A short prediction horizon. You only need to forecast ~10–20 seconds ahead, and only to the resolution of a wheel half — even predicting the correct region of the wheel converts a −2.7% game into a positive one.
3. The killer feature: you can bet after the ball is released. Croupiers traditionally call "no more bets" only after launch. That window — information arriving after the randomising process starts but before wagers close — is the entire trade. Every successful system, from Thorp's toe-switches to Small and Tse's timings, lives inside that window.

Why a lottery machine is a different universe

Now put a gravity-mix lottery machine next to that wheel and count the ways the problem explodes.

Many bodies, not one. Instead of one ball on a rim, a draw machine tumbles dozens of balls colliding with each other, with paddles, and with the chamber — for a 6-from-45 game, 45 interacting projectiles. Each collision multiplies sensitivity: a microscopic error in one ball's state redirects the next collision entirely. This is the textbook mechanism of chaos — Small and Tse's own paper frames even one ball's late bounces as the chaotic, unpredictable phase of roulette. A lottery machine is essentially built out of nothing but that phase.

Long mixing, no shortcut. Roulette prediction works because the predictable rolling phase dominates and the chaotic bouncing lasts a moment. Draw machines invert the ratio: balls are churned for an extended mixing period before any ball is captured, precisely so that no measurable initial condition survives into the outcome. Exponential error growth over many collision times means any initial measurement, however precise, is erased — the same reason weather forecasts die at a couple of weeks, compressed into seconds.

And the killer feature is absent. Here is the argument that ends the discussion regardless of how good your physics is — and it needs no citation, only logic. Roulette prediction profits from the gap between ball release and bets closed. In a lottery, that gap does not exist: your entry is locked in before the draw begins. Even a hypothetical demon who could film the machine and compute perfectly would learn the winning numbers after every ticket is already purchased. Physics prediction without a live betting window is a spectator sport. (An insider who could act on advance knowledge is a different crime with a different mechanism — that's the Eddie Tipton story, and it involved rigging software, not predicting balls.)

The honest conclusion

• Physical prediction of gambling devices is real science with real wins — Thorp and Shannon's +44% lab edge and Small and Tse's published +18% return deserve their legend.
• It works only where dynamics are simple, the horizon is short, and — decisively — where you may bet after the randomisation starts.
• Lottery ball machines are engineered at the opposite corner on all three axes: many-body chaotic mixing over long times, with all entries closed before the first ball moves. The prediction channel isn't merely narrow; it's structurally sealed.

The practical upshot mirrors the rest of this cluster. Draw machines are, functionally, true random number generators — physical chaos as an entropy source, with mixing time doing the job that cryptographic conditioning does in silicon. If you want to experience how featureless genuinely random draws look — streaks, gaps, near-misses and all — run a few thousand through our draw simulator; and if you want numbers of your own drawn with the digital equivalent of a well-mixed machine, that's exactly what our generator is built to be.

Roulette was beaten by physicists. The lottery isn't waiting for a smarter physicist — it's designed so there is nothing for one to grab.