Load a software roulette game, place a bet, watch the wheel spin. The animation isn’t deciding anything. The result was settled on a server before the first frame was rendered. What you are watching is an animation of a number the software already picked, not footage of anything that spun. GLI-19, the testing standard for interactive gaming systems, says “the software shall not contain any logic utilized to generate the result of any game,” and that game outcomes “shall be generated by the Gaming Platform and be independent of the Remote Player Device.” Your device is a display. The decision happened somewhere else, and it happened first.
What a Certified Generator Has to Do
The requirements are a lot less vague than the marketing that references them. GLI-19 wants each possible selection to be “equally likely to be chosen” and requires that “knowledge of the numbers chosen in one draw shall not provide information on the numbers that may be chosen in a future draw.” The generator has to be “cryptographically strong.” It “shall not be seeded from a time value alone.” And it has to keep altering its own state using external entropy, which the standard justifies as “limiting the effective duration of any potential exploit by a successful attacker.”
A test house reviews source code for the randomness, scaling and shuffling algorithms, then runs statistical tests on what the standard calls the final outcome output, meaning the pocket numbers after scaling rather than the raw stream. A flawless generator wrapped in a bad scaling function still produces a biased pocket number, which is why the statistical work is done on the numbers players actually see.
A separate rule governs what an operator may then do with the numbers. Britain’s technical standards require that received random numbers be used in the order they arrive and not discarded except as a logged error, so an operator can’t draw a result, look at it, and draw again. That still isn’t a guarantee. NIST, whose suite much of the industry leans on, states that “no set of statistical tests can absolutely certify a generator as appropriate for usage in a particular application, i.e., statistical testing cannot serve as a substitute for cryptanalysis.”
Which Version You’re Choosing
All of that applies to the software version, and none of it describes the other kind of table. Choosing to play live roulette means the number doesn’t exist anywhere until a ball stops moving. There’s no server-side draw to precede the animation, because there’s no animation. There’s a wheel, in a room, on camera, and a bet that has to close while the ball is still traveling. New Jersey writes that into its code for anyone streaming a table there: the system must “prevent anyone from accessing the wagering outcome prior to finalizing a wager,” and must record a dealer-verified result before posting it. The trade is that a certified process has been swapped for a physical object, and physical objects drift.
A Real Wheel Is a Maintenance Problem
Michael Small and Chi Kong Tse published a working model of the spin in 2012. While the ball is fast enough, the banked track supplies the inward force needed to hold it in a circle; below a critical speed set by the angle of that track it cannot, and the ball drops out of the rim. Time that moment and the landing zone narrows.
A tilted wheel narrows it a lot further. Small and Tse confirmed Edward Thorp’s claim that a tilt of 0.2 degrees is enough to make the ball systematically favor one half of the rim. The equipment industry knows the number. Interblock’s rim tilt detection patent, granted in 2024 as US11980823B2, describes an inclination sensor in the rim that measures down to 0.025 degrees, opens a warning band at 0.12 degrees and throws a hard error at 0.2.
The maintenance regime is published too. TCSJohnHuxley says its Saturn wheel takes 32 hours of work and more than 130 quality checks, that contact surfaces are measured with a laser device before dispatch, and that its drop zone detection monitors how often the ball leaves a particular area of the ball track. The documentation is blunt about why: an unlevel wheel “has the potential to generate a drop zone,” and rotating the wheel regularly reduces “the risk of operating with predictable drop zones.”
Drift has been worth real money to anyone able to measure it. Doyne Farmer’s account of the Eudaemons, the shoe-computer group that carried wearable machines into Nevada in the late 1970s, claims a 20 percent advantage over the house. Small and Tse measured 18 percent thirty-four years later, timing the ball by hand. Both results required a person standing at the wheel.
When the Bet Actually Closes
The last piece of a streamed table is the one part of it that gets no explanation on screen. “No more bets” is computed rather than called. Interblock’s game-cycle patent, granted in April 2020, describes a controller reading the ball’s launch velocity and its rate of deceleration and computing the bet close time from those figures, so the window shuts as late as the physics allows. The same filing gives fifteen ball revolutions as the target for a cycle.
On a software table the outcome is generated first and the standards govern how. On a live table it’s generated last, and what the operator sets instead is the moment your input stops counting. The two are audited by different people and they fail in different ways. What doesn’t change is that there’s a price on every spin either of them produces, and that the paytable and the count of green pockets set it. A second green pocket nearly doubles the house edge whether a croupier spins the wheel or a program does, and no amount of certification on either side moves that number a hundredth of a point.