From Thorp and Shannon's basement to the Cammegh RRS, why physics nearly broke the wheel.
Wheel physics and prediction
Annabel Cavendish
Editor · 14 May 2026
Shannon's Basement, June 1961
The beginning of the story is one of the genuinely interesting chapters in the history of applied mathematics.
Claude Shannon is one of the most important figures in the history of science. He invented information theory. He is the reason your phone can send data across a network. In the summer of 1961, he spent spare time in his Winchester, Massachusetts basement building a cigarette-pack-sized analog computer with 12 transistors, designed to time a roulette ball's revolution and predict which octant of the wheel it would land in. He built it with Edward Thorp, the mathematician who had already cracked blackjack card counting.
The device worked via a toe-operated microswitch in the wearer's shoe, which timed the ball, and a small earpiece that delivered one of eight tones corresponding to the predicted section. According to Thorp's 1998 paper hosted at the University of Virginia, the stated edge on the single most favoured octant under laboratory conditions was 44%. They tested it in Las Vegas that summer; the predictions were consistent with laboratory expectations, but hair-thin wires to the earpiece repeatedly broke, preventing any serious session. Thorp kept the project secret until 1966 and published the full technical account in a 1998 IEEE symposium paper. Guinness World Records credits it as the first wearable computer ever built.
The 44% figure gets repeated in popular accounts as though it was a general edge on every spin. It wasn't. That was the advantage on the best prediction in controlled conditions.
The Eudaemons: Twenty Percent in the Real World
The real-world version came from Doyne Farmer and colleagues at UC Santa Cruz in the late 1970s. They built a digital wearable into a shoe, hand-coded by Farmer in three kilobytes of machine language. As Farmer recounts on his website, they made over eleven trips to Las Vegas, Reno, and Tahoe. Their edge in casino conditions was approximately 20%.
Still extraordinary. Still not 44%. The hardware repeatedly failed, connections broke, and they never played for high stakes. The computer now sits on loan at the Heinz Nixdorf Museum in Paderborn. The broader account appears in Adam Kucharski's book The Perfect Bet, and the Wikipedia entry on the Eudaemons confirms the key details.
Farmer later found, incidentally, that air resistance rather than friction was the dominant force slowing the ball. This corrected a key assumption that later researchers had been working from.
Small and Tse, 2012: Physics-Based Prediction Still Works
The definitive modern proof that the physics works came in 2012, published in the AIP journal Chaos by Michael Small and Chi Kong Tse at Hong Kong Polytechnic.
Using a standard casino-grade wheel and a manual clicker device, they correctly predicted the half of the wheel 13 times in 22 trials, for an expected return of positive 18%. With a camera mounted above the wheel, results improved further. The critical finding: a tilt of just 0.2 degrees was, in their words, "more than sufficient" to introduce systematic, exploitable bias. According to the arXiv preprint of the paper, 0.2 degrees is an almost imperceptible lean.
The year was 2012. The iPhone 5 was released the same year. Physics-based roulette prediction was still being academically confirmed as effective while you were updating your operating system.
Why It Doesn't Work Now
Two reasons, and they are connected in a way that is almost elegant.
The TCS John Huxley Saturn wheel contains a patented inclinometer system. A patent filed in 2020, US11452934B2, describes sensors in the wheel rim that measure tilt to 0.025-degree resolution. The alert system fires a red error light when tilt exceeds 0.2 degrees. That is not a coincidence. Zero-point-two degrees is exactly the threshold at which Small and Tse's 2012 paper confirmed systematic bias becomes exploitable. The engineering of the Saturn wheel uses the same threshold as the physics research, which means the moment a wheel becomes exploitable, it displays an error and the game stops. The window between "exploitable" and "flagged" is essentially zero.
The Cammegh Mercury 360 RRS (Random Rotor Speed) system addresses the other vector of attack. Visual ballistics and computer-assisted prediction work by timing the ball's decaying orbit and predicting where it lands relative to the rotor. This requires knowing the rotor's speed with precision. The RRS system adds a random, imperceptible deceleration to the rotor at two points in each game: at ball launch, and when the ball drops below the no-more-bets threshold. The Cammegh 2025 brochure publishes an example showing approximately 7 rpm of additional variation caused by RRS in a single game. At that level of variation, any timing-based prediction of where the rotor will be when the ball lands becomes statistically useless, even with a computer.
The Ritz, 2004: The Last Win
The last documented case of a physics-based roulette win at a named London casino that was not subsequently reversed belongs to a player publicly known as Niko Tosa and two colleagues, who played roulette at the Ritz in March 2004 and left with approximately £1.3 million. Scotland Yard investigated; the press reported a laser scanner concealed in a mobile phone. As The Guardian reported in March 2004, the case raised serious questions about what constituted unlawful gaming devices under the then-current law. The three were ultimately not charged, keeping their winnings.
The UK government subsequently commissioned a review confirming basic roulette computers are effective, and the Gambling Act 2005's Section 42 broadened the cheating offence to cover any acts likely to affect the random element of a game. Given the Ivey standard now in effect, you don't have to be subjectively dishonest to have legally cheated. The window that existed in 2004 is closed.
The Historical Record
For perspective, here is where the physics-based advantage has stood across the decades.
Era / case
Method
Edge achieved
Outcome
Jagger, Monte Carlo, c.1880
Manual bias recording
Not quantified
~£80,000 withdrawn
Hibbs and Walford, Reno, 1947
Manual recording, pocket bias
Not quantified
~$8,300 from $300
Thorp and Shannon, Las Vegas, 1961
Analog wearable computer
+44% (best octant, lab)
Inconclusive due to hardware
Eudaemons, Las Vegas/Reno, 1978-1980s
Digital shoe computer
+20% (casino conditions)
Small winnings; never high stakes
Ritz team, London, 2004
Laser phone or visual sector tracking
Not published
~£1.3 million kept
Small and Tse research, 2012
Manual clicker, physics model
+18% (controlled)
Academic proof only
The story of wheel physics is one of genuinely brilliant people finding a real edge, and casino engineers being just as brilliant about engineering it away. The science is impeccable on both sides. On a modern Saturn or Mercury RRS wheel, the physics-based advantage is gone.
I'm Annabel, and this is the lesson where we talk about the most gloriously obsessive chapter in the history of gambling: the repeated attempts by mathematicians, physicists, and engineers to defeat roulette by treating it as a physics problem rather than a probability problem.
The short version is that it worked, several times, spectacularly.
The longer version explains why it doesn't work anymore, and the explanation is genuinely interesting.
Let's begin in June 1961, in the basement of Claude Shannon's house in Winchester, Massachusetts.
Claude Shannon, I should mention, is one of the most important figures in the history of mathematics.
He invented information theory.
He is the reason your phone can send data across a network.
In the summer of 1961, he spent his spare time building a cigarette-pack-sized analog computer with twelve transistors, designed to time a roulette ball's revolution and predict which octant of the wheel it would land in.
He built it with Edward Thorp, the mathematician who had already cracked blackjack card counting, and the device worked by means of a toe-operated microswitch in the wearer's shoe, which timed the ball, and a small earpiece that delivered one of eight tones corresponding to the predicted section.
They tested it in Las Vegas that summer.
The stated edge, on the single most favoured octant under laboratory conditions, was forty-four percent.
Thorp kept it secret until 1966, and the full technical account wasn't published until 1998 in an IEEE symposium paper.
Guinness World Records credits it as the first wearable computer ever built.
The forty-four percent figure gets repeated in popular accounts as though it was a general edge.
It wasn't.
That was the advantage on the best prediction in controlled conditions.
The real-world version is what Doyne Farmer and his colleagues at UC Santa Cruz achieved in the late nineteen seventies: a digital wearable built into a shoe, hand-coded by Farmer in three kilobytes of machine language, used on over eleven casino trips.
Their edge in casino conditions was approximately twenty percent.
Still extraordinary.
Still not the forty-four.
The hardware repeatedly failed, the wire connections broke, and they never played for high stakes.
The computer now sits on loan at the Heinz Nixdorf Museum in Paderborn.
Farmer later found, incidentally, that air resistance rather than friction was the dominant force slowing the ball, which corrected a key assumption that later researchers had been working from.
The definitive modern proof that the physics works came in 2012, published in the journal Chaos by Michael Small and Chi Kong Tse at Hong Kong Polytechnic.
Using a standard casino-grade wheel and a manual clicker device, they correctly predicted the half of the wheel thirteen times in twenty-two trials, for an expected return of positive eighteen percent.
With a camera mounted above the wheel the results improved further.
The critical finding was this: a tilt of just zero point two degrees was, in their words, "more than sufficient" to introduce systematic, exploitable bias.
Zero point two degrees.
That is an almost imperceptible lean.
The year was 2012.
The iPhone five was released the same year.
Physics-based roulette prediction was still being academically confirmed as effective while you were updating your operating system.
So why doesn't it work now?
Two reasons, and they are connected in a way that is almost elegant.
The TCS John Huxley Saturn wheel, which is one of the two dominant wheels in serious casino play, contains a patented inclinometer system.
The sensors measure tilt to zero point zero two five degrees of resolution.
The alert system fires a red error light when tilt exceeds zero point two degrees.
That is not a coincidence.
Zero point two degrees is exactly the threshold at which Small and Tse's 2012 paper confirmed systematic bias becomes exploitable.
The engineering of the Saturn wheel uses the same threshold as the physics research, which means the moment a wheel becomes exploitable, it displays an error and the game stops.
The window between "exploitable" and "flagged" is, on a Saturn-equipped table, essentially zero.
The Cammegh Mercury 360 with its Random Rotor Speed system addresses the other vector of attack.
Visual ballistics and computer-assisted prediction work by timing the ball's decaying orbit and predicting where it will land relative to the rotor.
This requires knowing the rotor's speed with some precision.
The RRS system adds a random, imperceptible deceleration to the rotor at two points in each game: at ball launch, and when the ball drops below the no-more-bets threshold.
The published brochure shows an example variation of approximately seven revolutions per minute of additional scatter in a single game.
At that level of variation, any timing-based prediction of where the rotor will be when the ball lands becomes statistically useless, even with a computer.
The rotor's final position is genuinely unpredictable.
The legal position in the UK is worth a sentence.
In Nevada, possession of a device to project gambling outcomes is a Category B felony.
In the UK, there is no equivalent specific statute, which is why Niko Tosa and his colleagues walked out of the Ritz in 2004 with approximately one point three million pounds and were ultimately not charged.
The Crown Prosecution Service found no applicable law at the time.
The UK government subsequently commissioned a review, confirmed that basic roulette computers were effective, and the Gambling Act 2005's Section 42 broadened the cheating offence to cover any acts likely to affect the random element of a game.
Given the Ivey standard now in effect, you don't have to be subjectively dishonest to have legally cheated.
The window that existed in 2004 is closed.
The story of wheel physics is one of genuinely brilliant people finding a real edge, and casino engineers being just as brilliant about engineering it away.
The science is impeccable on both sides.
On a modern Saturn or Mercury RRS wheel, the physics-based advantage is gone.
What remains is the house edge, working quietly and reliably at two point seven percent, exactly as it always has.
Don't let the romance of the physics tempt you.
The casino engineers read the same papers you did.
