Physics · the other kind of edge

Roulette wheel physics: the wheel has a body.

The payout table fixes the expected value of a fair wheel; no betting progression changes it. A physical wheel adds a separate question: could its motion or wear make outcomes measurably uneven? Here is the record, the measurement problem, and a model you can inspect in the simulator.

Two physical questions

Mechanical bias asks whether one wheel produces persistently uneven pocket frequencies. Physical prediction asks whether measurements of a single spin can narrow its likely landing sector. Both depend on the actual wheel and conditions. Neither follows from a Martingale, a streak, or a simulated hot number.

This distinction matters in the lab: a model can show how tilt changes a simulated distribution without proving that a casino wheel has the same tilt. A statistically unusual sample can also be chance, especially when you look for the hottest pocket only after collecting the spins.

Why this is a different question from "beating the math"

Betting systems (Martingale, D'Alembert, any progression you can write in Lua) operate on outcomes from a fair random process. Against a truly fair wheel, no sequencing of bet sizes changes the expected value; see Terms for why that's not up for debate here. Physical bias is a different mechanism entirely: it's not a strategy exploiting math, it's an observer exploiting a manufacturing defect or wear pattern in a specific, physical wheel. The math stays exactly as settled. The wheel doesn't.

The record

Joseph Jagger, Monte Carlo, 1873

Jagger, a Yorkshire mill engineer, paid clerks to record thousands of spins across six wheels at the Beaux-Arts casino. Five wheels showed nothing. One didn't: nine numbers came up conspicuously more than the rest. He bet those nine and won an amount reported in the tens of thousands of pounds before the casino, catching on, started rearranging wheels and fittings overnight to shake him off. He found the wheel anyway, by a scratch on its rim, and kept winning.

Gonzalo Garcia-Pelayo and family, Madrid, 1991–2004

A Spanish record producer organized his family to log spins by hand at Casino Gran Madrid, looking for statistically detectable bias: the same logic Jagger used, formalized. Over roughly a decade, across multiple casinos, the family won an estimated €1.5–2 million. The casino sued them for fraud. Spain's Supreme Court ruled in 2004 that recording public, observable information and betting on it is legal. There was no cheating, no device, no rule broken. Casinos worldwide subsequently adopted automated wheel-monitoring specifically in response.

The Eudaemons, Nevada casinos, late 1970s–80s

Physics graduate students J. Doyne Farmer and Norman Packard built one of the first wearable computers (hidden in a shoe, operated by toe switches) to time a spinning wheel and ball and predict the landing octant from measured deceleration, using nonlinear dynamics. Reported result: roughly a 20% edge over the house across multiple casino trips. This is single-spin physical prediction, not persistent bias-hunting: a different technique aimed at the same underlying fact: a real wheel obeys real physics, and real physics can be modeled.

The Ritz Casino, London, 2004

A trio used a laser rangefinder built into a mobile phone to clock ball and wheel speed, feeding a computer that predicted the landing sector, a modern, real-time descendant of the Eudaemons' approach. They won roughly £1.3 million over two nights. Police investigated for cheating; no charges were ever filed, because the relevant law only prohibited physically interfering with the equipment, and they hadn't: they had only observed it more precisely than the house expected anyone could.

It's in the peer-reviewed literature, too

This isn't only casino folklore. In 2012, physicists Michael Small and Chi Kong Tse published Predicting the outcome of roulette in Chaos, the American Institute of Physics' nonlinear-dynamics journal. Working from a casino-grade wheel and a simple hidden "clicker" to time the ball, they showed that a measurement of initial position and velocity is enough to predict the resulting half of the wheel well enough to turn the expected return from −2.7% into roughly +18% (Small & Tse, 2012). The house edge is arithmetic; the ball's trajectory is physics, and physics, it turns out, is measurable in real time. The full citation list lives on the Research page.

How many spins would reveal a biased wheel?

It depends on the size of the bias and how many pockets you inspect. Our wheel-bias power study calculates planning counts for a European wheel. It assumes a fair baseline of 1/37 per pocket, 80% power, and a 5% false-positive rate. The scan-all estimates correct for testing all 37 pockets; a pocket named before collecting data needs fewer observations.

True pocket excessNamed in advanceScan all 37
+0.25 points26,792 spins63,283 spins
+0.5 points6,887 spins16,110 spins
+1 point1,812 spins4,165 spins

These are normal-approximation planning estimates, not a verdict on a wheel. The collection rule, wheel identity, and stopping point need to be set before the result is examined. Conditions may also drift while spins are collected. Download the full table (CSV).

Test the model in 0xRoulette

The default outcome source in Play is seeded Mulberry32: independent, near-uniform pocket draws. Switch the outcome source to PHYSICS to use the deterministic motion model instead. In the lab, WHEEL.CONFIG exposes launch speed, friction, deflectors, and tilt; BATCH.SIM runs 1,000, 5,000, or 20,000 spins and plots counts in wheel order.

  1. Choose a European or American wheel, then select PHYSICS in the play controls.
  2. In the lab, leave tilt at zero and run a batch. Note the pocket counts.
  3. Increase tilt and run another batch. Compare the shape, then repeat before treating any peak as signal.

The model integrates wheel and ball motion with friction, a drop threshold, deflector interactions, and pocket capture. Each spin is reproducible for the same seed, wheel, and settings. The batch control currently generates a fresh seed when clicked, so its two runs are comparisons of distributions, not paired identical-seed trials. Its parameters are illustrative, not calibrated to any casino hardware. The simulator cannot diagnose a real wheel or predict its next spin.

What this means for 0xRoulette

The fair-wheel RNG and the configurable physics model answer different questions. The former is the baseline for betting-system backtests; the latter lets you explore how model assumptions change outcomes. Neither turns a simulated pattern into evidence about a particular physical wheel. The Research log and downloadable studies show the methods behind the published fair-wheel results.

Roulette physics, briefly answered

Does roulette wheel physics affect the outcome?

A physical ball and wheel follow motion, friction, and collisions, so their starting conditions can affect the landing sector. On a fair wheel whose starting conditions vary, that does not give a betting progression an edge. 0xRoulette offers a separate, configurable physics model for experiments; it is not a prediction of a casino wheel.

What is mechanical bias in roulette?

Mechanical bias is a persistent difference in landing frequencies caused by a particular wheel or its operating conditions. It must be measured on that wheel. A short run of hot numbers, or a result from this simulator, does not establish real-world bias.

How many spins does it take to detect a biased roulette wheel?

There is no universal number. In our European-wheel power calculation, a pocket with a true one-percentage-point excess over 1/37 needs about 4,165 spins when scanning all 37 pockets for 80% power at a 5% family-wide false-positive rate. A quarter-point excess needs about 63,283. These are planning estimates under stated assumptions, not proof from any single log.


Want the long versions? The Eudaemonic Pie, Beat the Dealer, and A Man for All Markets tell these stories from the inside; they're on our reading list, and buying through it keeps 0xRoulette free.

None of this is gambling advice, and it isn't a claim that real casinos today are beatable: modern wheel manufacturing and automated monitoring exist specifically because of the cases above. See Terms.