A casino random-number generator does not pay prizes directly. It produces or helps select unpredictable values, and the game software maps those values to cards, reel stops, dice totals or other outcomes. The paytable then converts the selected outcome into a credit or loss.
Fairness depends on the entire chain: entropy or seed generation, algorithm, state protection, unbiased mapping, game mathematics, transaction logging and controlled software changes. A statistically random-looking sample is not enough if another part of the system is wrong.
Most online games use pseudorandom generation
A pseudorandom number generator, or PRNG, is a deterministic algorithm. Starting from an internal state, it produces a sequence that appears random and passes defined statistical properties. If the complete state and algorithm were known, the sequence could theoretically be reproduced.
Security therefore depends on an unpredictable seed, sufficient internal state and protection against disclosure or manipulation. Operators can seed systems from operating-system entropy, hardware sources or combined inputs. Time alone is a weak seed if an attacker can narrow the possible value.
A true or hardware random-number source measures physical phenomena such as electronic noise. It can supply entropy directly or seed a cryptographic PRNG. Hardware output still needs health testing and secure integration; “physical” does not automatically mean flawless.
The game must map numbers without introducing bias
Suppose a generator produces integers from 0 to 99 and a game needs six equally likely dice faces. Simply taking the number modulo six is biased because 100 is not divisible by six: some remainders occur 17 times and others 16.
A correct implementation can reject values above 95 and redraw, leaving 96 values that divide evenly among six faces. Equivalent techniques can use larger ranges and carefully designed scaling.
Mapping becomes more complex in slots. The RNG may select a virtual stop on each reel. Multiple virtual stops can map to the same visible symbol, creating weighted outcomes. The animation then moves the displayed reels to the selected positions.
Counting visible symbols does not reveal probability unless the virtual mapping is known. A rare jackpot symbol can occupy one visible location while corresponding to very few virtual stops.
Randomness and payout percentage are different properties
An RNG can be perfectly random while the game has a high house edge. Roulette outcomes can be uniform and still return less than total stake because the payout is below fair odds. Slot reels can be independently selected and still produce a 90% or 97% theoretical return depending on the paytable and weights.
| Control question | What it addresses | What it does not establish |
|---|---|---|
| Are outputs unpredictable? | RNG security and state | Correct prize mapping |
| Are outcomes statistically distributed as intended? | Randomness and weighting | Low house edge |
| Does the paytable match the model? | Payment accuracy | Operator solvency |
| Is the certified version deployed? | Change control | That every account dispute is resolved correctly |
Return-to-player testing therefore combines RNG review with mathematical analysis and game simulation. The laboratory must verify both the selection mechanism and the economic result.
Independent outcomes do not compensate for losses
The UK Gambling Commission requires random outcomes to be acceptably random and prohibits adaptive or compensated behaviour in regulated remote games. A game should not lower a player’s future chance because the account recently won or increase it because the account recently lost.
Independence does not mean every short sample looks balanced. Ten reds can occur in roulette, and a slot can produce repeated low returns. Random sequences naturally contain clusters, gaps and streaks.
The gambler’s fallacy arises when a player expects the system to restore balance immediately. Long-run frequency describes a large population of trials, not a debt owed to one account.
Some bonus features use persistent meters or state, but the state should operate according to the published rules. A must-hit-by jackpot can become more likely near its ceiling without being secretly adjusted to a player’s personal history.
Shuffling cards requires a permutation, not repeated draws
Digital card games need an unbiased ordering of a finite deck. A common method is the Fisher–Yates shuffle, which swaps positions using random values so that every permutation is equally possible when implemented correctly.
Poor shuffling can arise if the random index range is wrong, modulo bias is introduced or cards are selected with replacement when the game requires a finite deck. A blackjack hand cannot receive the same physical card twice within one deck.
Persistent-shoe games must also preserve the undealt deck state securely between rounds. Games that reshuffle every hand should say so, because previous cards then contain no counting information.
Online poker adds peer-to-peer integrity: the deck must be hidden from players and unauthorized staff, and hand histories must allow later investigation without revealing live information.
Testing includes more than a statistical test suite
NIST’s SP 800-22 describes statistical tests for random and pseudorandom generators but explicitly warns that statistical testing cannot certify a generator for every application or replace design analysis. A predictable generator can sometimes pass frequency tests.
Gaming evaluation can include:
- algorithm and source-code review;
- seed and entropy assessment;
- state compromise and reseeding analysis;
- large-sample statistical testing;
- mapping and shuffle verification;
- game-mathematics and RTP calculation;
- error and interruption handling;
- transaction and audit-log review;
- security and change-management testing.
GLI-19 provides a widely used baseline for interactive gaming systems, while individual regulators can impose additional requirements. Certification applies to a defined version and configuration, not to every future software change automatically.
The result is normally selected before the animation finishes
In many RNG slots, dice games and digital wheels, the server selects the outcome when the wager is accepted. Reels, cards or a wheel then animate toward that recorded result. Pressing stop can shorten the display without changing the selected symbols unless the published game specifically includes player skill.
This architecture helps resolve disconnections. If the wager and result were recorded before the connection failed, the account can be settled from the server log even though the player did not see the animation.
A visual near-miss is therefore presentation, not evidence that the RNG almost selected a different prize. The game may deliberately display non-winning symbols near the payline, but regulated design and information rules determine what is permitted.
Live dealer games use different result sources
A live roulette wheel, physical card shoe or dice table derives the main result from physical equipment, not the same RNG used by a digital slot. Software still records bets, closes the market, recognizes the outcome and settles accounts.
RNGs can appear in supporting functions such as game-show bonus selection, random seat assignment or automatic shuffling, depending on the product. Those components require separate controls.
Video delay does not create a new result. The central record of the physical event and accepted wagers governs settlement. Rules should address invalid spins, misdeals, obscured cards and studio interruptions.
Provably fair systems solve a narrower problem
Cryptocurrency games may publish a hash commitment to a server seed before play and reveal the seed later. Combined with a client seed and nonce, the player can reproduce the result. This can show that a committed outcome sequence was not altered after the wager.
It does not prove that withdrawals will be paid, that the operator is solvent, that the house edge is correctly described or that the website follows local law. A regulated RNG and a provably fair system use different trust models, and each can fail outside its narrow proof.
Verification must use the published algorithm and the exact round data. Merely displaying the words “provably fair” is not evidence.
How a player can evaluate RNG claims
- Verify the operator and regulator covering the account.
- Identify the game supplier and exact game version where available.
- Read the rules, RTP and whether outcomes are independent or state-based.
- Check whether a recognized laboratory tested the relevant system.
- Do not treat short streaks as evidence of manipulation.
- Preserve bet IDs and account records when disputing a result.
- Separate game fairness from operator payment and complaint risk.
An RNG is one component in a controlled wagering system. The correct question is not whether numbers “look random,” but whether the generator, mapping, mathematics and deployed software together produce the approved probabilities and payments.
Official references include the UK Gambling Commission’s RNG standard, GLI-19 and NIST’s statistical test suite. Related GambleRoad guides cover slot RNGs, casino house edge and provably fair crash games.