What Is an RNG? How Random Number Generators Work

A random number generator, or RNG, is the part of a game that decides what happens: where the mines are, which number the roulette ball lands on, how high the multiplier climbs before it crashes. In software it is an algorithm that produces numbers which are, for every practical purpose, unpredictable and evenly spread. Every digital casino-style game depends on one, and the quality of that RNG is the difference between a fair game and a rigged one.

What an RNG does

A game needs random choices with exact probabilities. Roulette needs a number from 0 to 36 with each value equally likely. Mines needs 3 distinct tiles out of 25. Crash needs a multiplier where the chance of reaching x is 0.97 divided by x. Each of these is built from the same raw material: a stream of numbers that look uniformly random between 0 and 1.

The game then turns that stream into outcomes:

The conversion step matters as much as the generator. The UK regulator's technical standard says scaling must preserve the randomness (details below), because a careless "modulo" or "round" can quietly favour some outcomes.

Pseudo-random versus truly random

There are two families of random source:

• True random numbers come from physical noise: electrical jitter, radioactive decay, atmospheric noise. They cannot be predicted even in principle, but they are slow and need processing.
• Pseudo-random numbers come from an algorithm that is started from a small unpredictable number called a seed. Given the same seed, the algorithm produces the same stream. A well-designed algorithm seeded from a good source is computationally infeasible to predict.

"Pseudo" sounds like a weakness, but it is what almost every digital game uses, including regulated ones. The meaningful split is between ordinary pseudo-random generators (fine for animations or a screensaver, predictable if you know the state) and cryptographically secure ones, designed so that seeing a long run of outputs gives you no practical way to guess the next.

Where the draw happens in Cashout Arcade

All results are drawn on our servers, never on your phone. When you place a bet, the server draws the random numbers using the secure random source built into its runtime (the standard Web Crypto getRandomValues function). Two 32-bit values are combined into a 53-bit number between 0 and 1, which is the finest resolution a standard floating-point number can hold.

That means:

1. The result exists before you see anything. For an instant game such as Roulette, the server returns the finished result and the app plays it back. For a round game such as Mines, the mine positions are drawn when the round starts and stay on the server until the round ends.
2. The screen cannot change the outcome. The odds page puts it plainly: "Every result is decided on our servers." Anything you tap during an animation is a request for the next step, not an input to the draw. Gift Wall boxes and Lucky Pick cards are the clearest examples: the result is decided before you tap, so the box or card you choose does not change it.
3. The Crash curve does not leak. The crash point is drawn at the start and held on the server. The server does not reveal it until a cash-out could no longer succeed, so polling the round cannot tell you when it will end.
4. Skill games are checked on the server too. In Shell Game the server decides every swap and checks your tap against where the ball really ended up.

See are casino games rigged? for how this fits with the published odds, and the Odds page for the rules of each game.

What good looks like: the regulator's standard

You do not have to take "random" on trust, because regulated markets spell out what it means. The UK Gambling Commission's RTS 7 says outcomes must be "acceptably random", meaning you can show with high confidence, using generally accepted statistical tests, that they are. The guidance lists what to expect:

• Uniform distribution across the range, with outcomes matching the theoretical probabilities of the game.
• Unpredictability: for a software RNG, predicting the next number without knowing the algorithm and seed should be computationally infeasible.
• No cycling or synchronisation: the stream should not repeat, and two instances should not produce the same stream.
• Careful seeding: seeding or re-seeding must not introduce predictability.
• Scaling that preserves all of the above.

RTS 7 also says games must follow their published rules and payouts, random numbers must be used in the order received, adaptive or compensated games (ones that change their odds to correct a player's run of luck) are not allowed, and the display must show the result clearly. This standard applies to UK-licensed real-money operators. Cashout Arcade is not one, but it is a useful benchmark for what an RNG should deliver. Certification by an approved test lab is how regulated operators prove they meet it.

What we have and have not proved

• We publish the odds of every game, so the target is checkable.
• Our automated tests run seeded simulations through the real game code and compare the results with the exact figures, and a separate test holds the in-app odds document against the game engines.
• We have not published an independent lab certificate, so we do not claim one. A published certificate would only tell you the lab saw a particular version of the software.
• Our games are not "provably fair" in the cryptographic sense: you cannot verify an individual round after the fact. The provably fair guide explains what that scheme proves and why it is a different model from ours.

What you can check yourself

A simple frequency check will not prove a game fair, but it can flag a game that is far off its published chance. Pick a bet with a clear published probability, record the results, and compare:

• Crash with an auto cash-out at 2.00x should reach 2x about 48.5% of the time.
• A Roulette bet on red wins 18 of 37, or 48.65%.
• A Ball Drop drop on 12 rows lands in an edge bucket 1 time in 4,096 (each side).

Random results scatter around the expected frequency. As a rule of thumb, about 95% of honest samples land within two standard deviations of the expected rate. For a bet that wins 48.5% of the time, that band is:

If your record falls inside the band, there is nothing to see. If it falls outside, that is not proof of anything either, because 1 sample in 20 will do so by chance. It is a prompt to record more rounds. Anything under a few hundred rounds is almost always noise.

Two traps to avoid when checking: do not stop recording the moment the pattern looks bad (that is selecting your sample), and do not mix bets with different probabilities in one tally.

In short

• An RNG turns an unpredictable source into outcomes with exact probabilities.
• Cashout Arcade draws every result on the server with a cryptographically secure source, and the app only displays it.
• The 97% figures and every payout table are published, so you can compare what you see with what is claimed.
• Randomness shows up as scatter. Streaks and runs are part of it; see streak myths.