How RNGs are tested before and after an online slot goes live
Online casino RNGs are checked by Gambling Commission-approved test houses before release, through code review and statistical testing, then monitored once games go live.
Add as a preferred source on Google
Before a random number generator (RNG) can be used in an online casino game offered by a Gambling Commission licensee, an independent test house approved by the Commission has to test it. The test house reviews how the RNG is built, checks its code, runs statistical tests on what it produces and reports the results. That report has to reach the Commission before the game or RNG is released, and the checking continues once the game is live.
An RNG is the software that produces the numbers a game turns into results. What follows covers how that software, and the games built on it, are checked.
Who does the testing?
The Gambling Commission keeps and publishes a list of approved test houses authorised to carry out third-party testing. The licensee and its chosen test house agree the scope of the work, and the Commission requires that scope to be wide enough to properly assess compliance with its technical standards for online gambling. Remote casino licences are among those that need independent testing of games and RNGs.
What does an RNG test involve?
The Commission sets out four main strands:
- Documentation review. The tester reads the RNG documentation to understand how the RNG is used in the gaming system.
- Research. The tester looks into the RNG's algorithm or hardware for any publicly known weaknesses or vulnerabilities.
- Source code review. The tester checks that the code does what the documentation says it does.
- Statistical testing. The raw output of the RNG is tested, along with its output after it has been scaled to fit a game or used to shuffle virtual card decks.
Any problems are reported to the supplier and re-evaluated once fixed. The RNG report records who tested it and when, a certificate reference, a description and version of the RNG, the tests applied, the results and any limitations on how the RNG can be used.
What counts as random enough?
The Commission's standard is that RNG output and game results must be "acceptably random". That means it can be shown to a high degree of confidence, for example through generally accepted statistical tests, that they are random. In practice the RNG should be able to show that:
- its output is spread evenly across its whole range, and game outcomes occur in line with their theoretical probabilities
- the next number can't be predicted without complete knowledge of the algorithm and its seed, the starting value it works from
- it doesn't fall into repeating the same stream of numbers, and two copies of it don't produce the same stream
- seeding and re-seeding don't make it predictable
- scaling the output to fit a game keeps all of these qualities
The same standard rules out adaptive behaviour, also called compensation. An online game can't change the probability of outcomes during play to steer itself towards a target return.
Is the game tested as well as the RNG?
Yes. A sound RNG isn't enough if the game uses its numbers badly, so the test house also verifies the game design: its maths, the rules and artwork shown to players, and its theoretical RTP, the share of stakes the game is designed to pay back as prizes on average. It then tests the software in an environment that reflects the live one, including how raw RNG numbers are converted into game outcomes. Three methods are used:
- Simulation: running the game automatically over a large number of games to check the actual RTP falls within an acceptable range of the expected figure
- Emulation: recreating rare outcomes, such as jackpot triggers, special features and maximum prizes, to check they work
- Manual play: playing the game to confirm that what appears on screen, such as pay lines, behaves as expected
These checks connect to other online rules. Random numbers must be used in the order they arrive rather than discarded to change outcomes, results must follow the probabilities and pay tables described to players, and a game can't show a near miss in place of a different losing result.
A worked example
Here's a hypothetical illustration of why statistical testing needs volume. Say an RNG's output has been scaled to pick one of 10 equally likely numbers, 1 to 10. Over 1,000,000 draws, each number should appear close to 100,000 times, which is 10% of the draws. Over only 100 draws, one number turning up 14 times and another 6 times would be ordinary random variation, not evidence of a fault. That's why testing relies on large samples and judges results against a tolerance rather than expecting exact counts.
The same thinking applies to RTP. Say a game has a stated RTP of 96% and a simulation run records £1,000,000 staked and £958,000 paid out in prizes. The actual RTP for that run is £958,000 divided by £1,000,000, or 95.8%. Whether that's acceptable depends on how volatile the game is and how many games were played. The Commission describes the allowed tolerance as wide when only a little play has been measured, narrowing as play builds up.
What happens after a game goes live?
Testing before release isn't the end of it. The Commission requires online operators to monitor how their games perform, because a design, implementation or operational issue can slip past testing and leave a game overpaying or underpaying. Operators compare the actual return, worked out by dividing total prizes by total stakes, with the theoretical RTP and raise alerts when a game appears to fall outside its acceptable range. An error confirmed by investigation has to be reported to the Commission.
The Commission's guidance is realistic about chance here. Its worked tolerance uses a 95% confidence level, which means a game working properly could still fall outside the range in roughly 1 in 20 measurements. One failed check doesn't prove a fault; repeated failures might.
Updates are controlled too. A change that affects game fairness needs external retesting and a new report before release. A minor update that doesn't affect fairness can go out without retesting, but it must be recorded under change control, with a named personal management licence holder responsible. If the underlying RNG or game platform changes, a representative sample of games has to be retested. An approved test house audits these processes every year, and the audit includes a review of how RTP is monitored.
Live dealer games work differently, because results come from physical equipment such as roulette wheels and cards. There the focus is on controls over the equipment, shuffling and dealing, although results can still be measured over a long period to check they are acceptably random.
What does testing tell you, and what doesn't it?
The way I'd put it: testing confirms that a game behaves the way it says it does. It checks the RNG is unpredictable, that outcomes follow the stated probabilities and that the rules and RTP match the software. The Commission also requires the rules, and information such as the RTP, house edge or probability of winning, to be easily available before you commit to play.
Testing doesn't make results any less random, and it doesn't change the game's design. Where a game's RTP is below 100%, the difference between what is staked and what comes back as prizes is kept by the operator on average over a very large amount of play. Individual sessions can land well above or below that average.
The practical point to keep: a tested RNG means each result is random and the game works as described, not that wins become more likely or easier to predict. Any result can be a loss, and over time playing costs money.
Sources: UK Gambling Commission, Testing strategy for compliance with the remote technical standards: procedure for testing; UK Gambling Commission, Remote gambling and software technical standards: RTS 7, generation of random outcomes; UK Gambling Commission, Live return to player performance monitoring of games of chance; UK Gambling Commission, How to calculate return to player (RTP); UK Gambling Commission, Remote gambling and software technical standards: RTS 5, result determination; UK Gambling Commission, Remote gambling and software technical standards: RTS 3, rules, game descriptions and the likelihood of winning.
How this was written. Drafted with AI assistance from the sources listed above in Harvey Watkins's style and published after automatic checks for accuracy and compliance, without a line-by-line read by an editor. If you spot a mistake, tell us. Our editorial standards.


