How the RNG actually works inside a certified pokie

What a certified gaming RNG actually is

A random number generator, in the context of a certified pokie, is a deterministic algorithm seeded from a hardware entropy source that produces a sequence of integers indistinguishable from truly random across every practical statistical test. The output is drawn on demand, one integer per reel stop determination, and it is not retained beyond the current spin.

The word deterministic is important. The algorithm is a pure function, which means given the same seed and the same request sequence it produces the same output. What prevents an attacker from reproducing the output is that the seed is refreshed continuously from a hardware source that cannot be observed or replayed.

Two properties matter for the player. The output is unpredictable to any practical adversary, including one with source-code access. And the output is independent between spins, so past outcomes carry no information about future outcomes. Both properties are hard requirements of the certification regime.

The hardware entropy source

The entropy source is where the true randomness enters the system. Modern gaming cabinets use one of three common designs.

Ring oscillator arrays. Small unstable oscillator circuits produce jitter that is sampled at high frequency, xored together and processed into a stream of high-entropy bits. Cheap, reliable, widely used in consumer and gaming hardware.

Avalanche diode reference circuits. A reverse-biased diode operated at its breakdown voltage produces thermal-noise-driven current variation, amplified and digitised into an entropy stream. Historically favoured in high-end cabinet designs.

Silicon thermal noise sensors. Purpose-built silicon designs harvest thermal noise from a resistor or a small analogue circuit and stream digitised bits. Common in modern smartphones and increasingly in gaming hardware.

All three designs pass certification if properly implemented and monitored. Certification labs check the entropy source's health indicators, its statistical properties and the health-monitoring code that would halt the system if the source drifts out of specification.

The pseudo-random algorithm and modern choices

The entropy source drives the seed. The pseudo-random algorithm expands the seed into a long deterministic sequence usable for many reel determinations before the seed is refreshed. The algorithm is where design decisions differ between studios.

ChaCha20 is a cryptographic stream cipher widely used in modern gaming RNGs. It has strong statistical properties, resistance to state recovery attacks and a well-understood security proof.

xoshiro256++ and xoshiro256** are non-cryptographic PRNGs designed by Blackman and Vigna, favoured for their speed and their strong equidistribution properties. They are common in newer online-native studio implementations.

AES-CTR mode constructions, using AES in counter mode with a rekey schedule, are the third common family. Their advantage is that the cryptographic strength is well understood and the implementation can share hardware acceleration with other cryptographic uses in the cabinet.

Older studios may still use variants of Mersenne Twister, which passes most statistical tests but has known cryptographic weaknesses. Where Mersenne Twister is used, the certification regime requires additional mitigation, typically frequent reseeding from the entropy source.

Seeding and continuous reseeding

The initial seed for a gaming RNG is drawn from the hardware entropy source at boot and used to initialise the pseudo-random algorithm's state. Certification requires that this initial seed contain enough entropy that no practical adversary can guess it, typically at least 256 bits of entropy.

Continuous reseeding is what maintains the unpredictability guarantee across the machine's operational lifetime. Every so often, typically after a fixed number of draws or on a time interval, the algorithm's state is mixed with fresh entropy from the hardware source. This is called reseeding and it prevents state-recovery attacks that might otherwise succeed given a very long output sample.

Reseeding rates vary by design. Some cabinets reseed on every draw, which is conservative and slow. Others reseed every hundred or thousand draws, which is fast and still secure under standard cryptographic assumptions. Certification labs check the reseeding rate against the chosen algorithm's known state-recovery bounds.

The draw per spin and reel mapping

When the player presses the spin button, the cabinet code requests a sequence of integers from the RNG, one per reel. Each integer is in the range zero to virtual-strip-length minus one, drawn from a uniform distribution.

The virtual strip is the underlying reel design. Typical modern pokies use virtual strips of one hundred and twenty eight, two hundred and fifty six or five hundred and twelve positions. Each position on the strip maps to a specific visible symbol, with high-value symbols appearing rarely and filler symbols appearing often.

Once the integer is drawn, the mapping is deterministic. The cabinet code looks up the position on the virtual strip, resolves it to a visible symbol, and adds it to the reel window for that spin. The paytable then evaluates the visible window across all paylines or Ways-to-Win positions and computes the payout.

The whole process, from button press to payout display, typically runs in under fifty milliseconds. The reel animation that follows is presentation, not computation. The outcome was determined at the instant of the spin request.

The independence property and its implications

Independence between spins is a hard property of the RNG design. Every spin draws from the same underlying distribution, and the outcome of any one spin carries no predictive information about the next.

This is the property that makes the due-for-a-payout intuition wrong. A machine that has not paid out in a long time is not more likely to pay out on the next spin. A machine that just paid a jackpot is not more likely to go cold. The RNG has no memory of the session.

Independence is verified in certification through auto-correlation tests, runs tests, gap tests and higher-order statistical tests. A pass means that no computable pattern relates the outcome of one spin to any sequence of previous outcomes.

Practically, this means every session starts fresh in the sense that the RNG's future output does not depend on the player's history. A player who lost their first ten spins is not owed a win on the eleventh. A player who won their first ten is not due a cold streak. Every spin is an independent draw.

Stateless with respect to the player

A subtly different property from independence between spins is that the RNG is stateless with respect to the player. The RNG does not know how much you have deposited, how long you have been playing, what your session bankroll is or what your betting pattern has looked like.

This matters because it rules out entire families of conspiracy theory. The RNG cannot adapt to your bankroll. It cannot decide to punish you for cashing out early or reward you for staying longer. The paytable configuration served by the operator can differ between titles or between operators, but within a single title on a single operator the RNG behaviour is uniform across all players and all sessions.

What can change session to session is your subjective experience of the same underlying distribution. A run of unlucky spins is a genuine draw from the tail of the distribution, and it is not evidence that the machine is behaving differently for you. Statistical thinking here is the correct engineering habit.

Certification testing and the statistical batteries

Certification testing runs the RNG output through a battery of statistical tests. The three main batteries used in the gaming industry.

The NIST Statistical Test Suite is a federal standard from NIST that includes fifteen individual tests covering frequency, block-frequency, runs, longest run, matrix rank, spectral, non-overlapping template, overlapping template, universal, linear complexity, serial, approximate entropy, cumulative sums, random excursions and random excursions variant.

Diehard and its successor Dieharder run a further set of tests including birthday spacings, overlapping permutations, ranks, monkey tests, count-the-ones tests, parking lot, minimum distance, random spheres, squeeze, sums, runs and craps.

TestU01, developed at the University of Montreal, is the modern reference and includes a battery called BigCrush of one hundred and six individual tests. Passing BigCrush is a high bar and only cryptographically strong RNGs typically pass it.

Gaming-specific tests supplement these with checks on period length, seed reproducibility, reseeding effectiveness and hardware entropy source health monitoring. A certified pokie has passed all of the above.

Common attacks and the defences that stop them

Historically, several gaming RNG attacks have succeeded on poorly designed systems, and the defences against them are worth naming so you know what a certified modern design is protecting against.

State-recovery attacks. An attacker observes a long sample of RNG output and tries to reconstruct the internal state, then predicts future outputs. Modern defences use cryptographically strong PRNGs (ChaCha20, AES-CTR) that resist state recovery even given very large output samples, combined with continuous reseeding.

Seed prediction attacks. An attacker guesses the initial seed and reproduces the output. Modern defences use at least 256 bits of entropy in the initial seed and continuous reseeding from a hardware source.

Side-channel attacks. An attacker observes power consumption, electromagnetic emissions or timing patterns and infers RNG state. Modern defences include shielded hardware, timing-neutral implementations and, in high-assurance designs, redundant computation.

Malicious firmware attacks. An attacker replaces the cabinet firmware with a version that biases the RNG or leaks state. Modern defences include signed firmware, hardware-attested boot chains and regulatory sealed-cabinet requirements.

Misconceptions and engineering corrections

Five common misconceptions about gaming RNGs, each with the engineering correction. As a repeatable habit for evaluating RNG claims, the following verification steps are worth applying.

1. 1Check the certification badge. Look for a lab identifier (GLI, iTech Labs, BMM or eCOGRA) in the game info panel or the operator footer.
2. 2Verify the lab's own register. Every mainstream certification lab publishes a searchable register of certified titles. Verify the specific title you play.
3. 3Read the study reference. Certified titles publish a study reference date. Older than three years and it may be worth checking whether the operator serves the current release.

Misconception one, the machine is warmer in the evening or on weekends. The RNG has no clock-dependence in its output distribution. Warmer sessions are variance, not schedule.

Misconception two, a specific machine or seat is lucky. Independence between spins means the machine has no memory. Human perception clusters on recent outcomes and misreads variance as personality.

Misconception three, if I press the button at a specific rhythm I can influence the outcome. The outcome is determined at the instant of the spin request. Rhythm changes only the timing of that request, not the outcome distribution.

Misconception four, autoplay behaves differently than manual play. It does not. Autoplay requests spins at fixed intervals; the RNG treats each request identically to a manual one.

Misconception five, the cabinet operator can tilt the RNG in favour of the house at will. In a certified cabinet the RNG is sealed and cannot be tilted at run time. The operator's degrees of freedom are limited to which certified paytable configuration is served.

Cabinet floor RNG versus online server RNG

A land-based cabinet runs the RNG locally, inside sealed hardware, with the entropy source physically inside the cabinet and the algorithm running on an embedded processor. This design is regulated at the state level and inspected physically.

An online pokie runs the RNG on a server, in a data centre operated by the studio or by a game aggregator. The player's browser communicates with the server; the server draws the RNG output and returns the reel positions. This design is regulated by the studio's certification lab and audited through logs rather than physical inspection.

Both designs pass the same statistical batteries. The difference is in the trust model. A cabinet's sealed hardware and physical regulatory inspection is a stronger anti-tampering guarantee than a server's log audit. In practice the trust gap is small on any well-certified studio, and offshore online titles from major studios are engineered to a comparable RNG standard.

Where the gap opens is on the operator conduct side, which is a separate question from RNG certification. Certified RNG on an untrustworthy operator is not safer than uncertified RNG on a well-run operator would be. In practice the well-run operators universally serve certified games, so this concern is theoretical rather than live.

Responsible play in the context of RNG certainty

Understanding the RNG changes the framing of the play. A certified RNG on a well-designed pokie produces outcomes indistinguishable from truly random across every practical test. There is no exploit, no pattern, no seat that is luckier than any other, no rhythm that beats the maths.

What varies is your session outcome, which is a draw from a fixed distribution. You cannot influence the draw. You can influence which distribution you play against (title selection), how many draws you take (session length), how much each draw costs (stake size) and when you stop (walk-away discipline).

Those four levers are the whole game. Once you have set them, the RNG does its work, and your session outcome falls where variance places it. The engineering perspective is that this framing takes some of the anxiety out of a losing session and some of the mysticism out of a winning one.

If your play stops feeling recreational, or if you find yourself believing the machine owes you a payout, that belief is the signal that matters. GambleAware is on 1800 858 858, twenty four hours a day, free and confidential. Gambling Help Online at gamblinghelponline.org.au offers web chat. Lifeline on 13 11 14. Beyond Blue on 1300 22 4636. The IGA 2001 prohibitions attach to operators, not players, and this article is educational rather than legal advice.