Provably Fair Casino Games: How the Two Methods Work, Round by Round
A player who loses a hand and cannot check the result has one option: trust you. A player who can recompute that same result on their own machine does not need to. That single difference is what provably fair casino games buy an operator. Every disputed spin stops being an argument and becomes a calculation.
Provably fair is not a certification, a badge or a marketing line. It is a cryptographic method that lets a player verify, after the round, that the outcome was fixed before they bet and was not changed to make them lose. This guide explains how the method works, the two models our game engine runs and the exact arithmetic behind each, what a verification looks like on a higher-or-lower card game and on a poker deck, and which of our 251+ HTML5 games carry it. It sits on top of a GLI-19 certified RNG; it changes nothing about the mathematics or the return to player, and everything about how much a player has to take on faith.
How provably fair actually works
Every provably fair round is a sealed handshake built from three ingredients:
- The committed value: a secret the game fixes before the round and keeps hidden while it runs. Depending on the model, this is either the outcome itself or the seed the outcome will be derived from.
- The client seed: a number the player controls and can change at any time. Because the player contributes input the house cannot predict, the house cannot pre-arrange a losing result.
- The nonce: a counter that moves on every round, so the same pair of values never produces the same outcome twice.
- The fingerprint: the SHA-256 hash of the committed value, published before the bet. It is the only thing the player sees until the round is over, and it is enough.
The sequence is always the same, and the order is the whole point:
- Commit. Before the player bets, the game publishes a SHA-256 fingerprint of the hidden value. The fingerprint gives nothing away, because the function cannot be run backwards, but it locks the value in. It is a sealed envelope held in escrow.
- Play. The player sets a client seed and plays the round. The outcome is derived from the committed value plus that seed.
- Reveal. When the round ends, the game reveals the hidden value.
- Verify. The player hashes the revealed value. If it matches the fingerprint published in step one, nothing was swapped, and they can re-run the formula to confirm the result they saw was the result the mathematics produced.
No certificate can do this. A laboratory audit tells a player the random number generator is sound in general; it can say nothing about one specific hand they think was rigged. Provably fair answers that exact question, per round, without anyone having to take the operator's word for it.
The two models our engine runs
"Provably fair" is a principle, not a single algorithm. Players who came from one ecosystem expect the verification flow they already know, so our games can run either of two established models. The operator chooses one per installation. Here is what each one does, step by step, because the detail is what a technical buyer will ask for.
Model 1: the committed outcome
This is the default. The certified random number generator draws the entire outcome of the round first: the reel stops, the roulette pocket, the shuffled deck. The game then builds a salt from 32 random bytes plus a draw from the same generator, and publishes the fingerprint SHA-256(outcome | salt) before the player bets.
The player's seed, a number between 0 and 999,999,999, is added to the nonce and shifts each committed value within its own range, so a reel stop or a pocket moves by an amount the house could not know when it sealed the envelope. After the round the game reveals the outcome and the salt. The player hashes the two together, compares the result with the fingerprint, and applies the same shift to see the result they were paid on.
The strength of this model is the simplicity of the check: one published hash, one revealed pair of values, one comparison. For an audience that is not crypto-native, it is the easier story to tell and the easier proof to follow.
Model 2: the committed server seed, with HMAC
This is the method the largest crypto casinos made standard, and it is the one a crypto player will expect. The game generates a 64-character hexadecimal server seed and publishes SHA-256(server seed) before the round. The outcome is then derived from the seed in the open:
- The game computes HMAC-SHA256 with the server seed as the key and the string "client seed : nonce : block" as the message, as defined in RFC 2104.
- The 32 bytes that come out are read four at a time, and each group becomes a number between 0 and 1: byte one divided by 256, plus byte two divided by 256 squared, plus byte three divided by 256 cubed, plus byte four divided by 256 to the fourth. One block yields eight such numbers; when a game needs more, the block counter moves on.
- Each number is multiplied by the size of the range it has to fill and rounded down: a reel of 50 positions, a wheel of 37 pockets, a deck of 52 cards.
After the round the game reveals the raw server seed, and the player can reproduce every number from scratch with any standard cryptographic library or an independent verifier. They do not have to trust a calculator the operator wrote. If your audience comes from crypto casinos, this is the model they will look for by name.
One detail that separates a real implementation from a demonstration: the committed data of a round that is still open is encrypted at rest on the game server. A database read during the round shows nothing a dishonest insider could use.
A worked check on a higher-or-lower card game
The easiest place to see the method is a game with one decision. Take Over or Under 26, our four-card joker game with a 96.60% RTP: the player is dealt cards and bets whether the total lands over or under 26. On the committed-outcome model the round runs like this.
Before the bet, the game draws the cards with the certified generator, builds the salt, and shows the player a fingerprint of the draw and the salt together. The player types in a seed of their own, or lets the game pick one, and chooses over or under. The cards are turned. After the round the game reveals the draw and the salt; the player pastes both into a SHA-256 tool, sees the published fingerprint come out, and knows the cards were on the table before they chose a side.
That is the whole of a provably fair hi-lo check: one hash before, two values after, one comparison. It takes a player under a minute, and most never do it. The ones who do are the ones who post about it, which is why it matters.
Provably fair poker, and how a deck is shuffled
Card games raise a harder question than a single draw: how do you prove a whole deck was shuffled honestly? On the HMAC model, the shuffle is a Fisher-Yates walk through the deck that consumes one number from the HMAC stream per swap. Fifty-one numbers shuffle fifty-two cards, and because each number comes from the committed seed, the entire order of the deck is fixed, and verifiable, before the first card is dealt.
This is what makes provably fair poker possible in our catalogue. On Oasis Poker (98.92% RTP), Pai Gow Poker and Red Dog, the dealer's hand and the player's hand come from the same committed deck, so neither can be adjusted after the player has acted. On the video poker titles, Jacks or Better (98.39% RTP) among them, the draw cards are the next cards of that same fixed order, which closes the oldest complaint in video poker: that the replacement cards were chosen after the hold.
A slot works the same way on both models. Each reel stop is one committed value, and a game that draws extra numbers after the stops, for a feature or a multiplier, keeps a ledger of those draws with their position in the stream so a player can follow every one.
Which games are provably fair
Provably fair applies to every game type in our catalogue except scratch cards. Counted in October 2026, that is 159 of our active games, across:
- Slots: the spin outcome is committed and revealable.
- Table and card games: blackjack, baccarat, the poker variants, Over or Under 26, and the war and dragon-tiger style games, where the dealt deck is verifiable against the committed value.
- Video poker, roulette and keno: the draw or the wheel result reproduces from the committed value. A keno draw of 20 numbers from 80 is one of the clearest cases, because the whole draw is on screen.
- Dice and crypto-native originals: Dice CyberWorld, Mines, Plinko, Limbo and Sic Bo, the set crypto players verify most often.
Scratch cards are the deliberate exception. They draw from a pre-generated, certified set of tickets rather than computing an outcome at the moment of play, and exposing the data needed to verify a ticket would also expose which other tickets are winners. So the 86 scratch titles stay on the certified ticket-set model by design, which is the right call for that format, not a gap. It is worth being precise about this with buyers, because "all our games are provably fair" is the kind of overclaim a serious operator will test.
Why operators should care, not just players
In crypto and sweepstakes markets, the single most common objection a new player raises is some version of "how do I know this isn't rigged?" Trust is the conversion bottleneck, and a certificate in the footer does not resolve it, because players cannot interrogate it. A result they can recompute themselves does. It moves the burden of proof off your support team and onto arithmetic that is true whether or not anyone trusts you.
There is also a defensive benefit. When a player disputes a result, "here is the fingerprint, here is the revealed value, here is the formula, run it" ends the conversation in your favour every time you are honest. It is cheaper than a support argument and more credible than a policy page.
Provably fair versus certified RNG: you want both
These are often confused, so it is worth separating them. A certified RNG is an independent laboratory audit, against the GLI-19 standard in our case, proving the random number generator is statistically sound and unpredictable. Provably fair is a per-round, self-serve proof that one specific outcome was committed in advance and not tampered with. The certificate establishes that the engine is trustworthy in general; provably fair lets a player confirm a single result with their own hands. Strong operators offer both, and they are careful to say "GLI-19 certified RNG", because it is the RNG that is certified, not each individual game.
One consequence most buyers do not expect: the two layers are linked at certification time. The fairness model is part of how an outcome is produced, so a game certified on one model that is switched to the other on a live installation is a change the laboratory would need to look at again. That is why the model is chosen per installation, before launch, and left alone. Switching is possible; it is simply not free of consequence, and we say so rather than let a buyer find out from a regulator.
What to verify before you offer provably fair games
If you are evaluating any provider's claim, ours included, these are the checks that separate real implementations from theatre:
- Is the commitment published before the bet? A hash revealed only after play proves nothing. The seal has to come first.
- Can the player control the client seed? If the house controls every input, it can grind for a favourable server seed. Player-supplied input is what closes that door.
- Does the committed value change every round? A fresh commitment per round, or a clean reveal-and-rotate cycle, keeps each outcome independently verifiable.
- Is there a public verifier? Players should be able to reproduce the result with an independent tool, not only a calculator the operator wrote and controls.
- Is the open round protected on the server? Ask what a database read shows while a round is in progress. The answer should be: nothing usable.
- Is the RTP unchanged? The fairness layer should sit on top of the published mathematics, which you can check against the RTP list and the PAR sheet of the game.
Our games meet all six, on both models, for every game type except scratch cards.
Frequently asked questions
What does "provably fair" mean in a casino game?
The result of a round is fixed before the player bets, a cryptographic fingerprint of it is published, and after the round the player can recompute the result and match it against the fingerprint. The player checks the fairness of their own round rather than trusting the operator or a certificate.
What is the difference between the two models?
The committed-outcome model seals the finished outcome and reveals it with its salt after the round; the check is one hash comparison. The committed-seed model seals a server seed and derives the outcome from it with HMAC-SHA256 in the open; the check reproduces every number from the seed. Crypto players usually expect the second; everyone else finds the first easier.
Is a provably fair hi-lo game different from a normal one?
The game plays the same. The difference is that the cards are drawn and sealed before the player chooses over or under, and the draw and its salt are revealed afterwards, so the player can prove the cards were not picked after the bet.
Can poker be provably fair?
Yes. The deck order is derived from the committed seed with a Fisher-Yates shuffle, one number from the HMAC stream per swap, so the whole deck is fixed and verifiable before the first card is dealt. Dealer hands, player hands and video poker draw cards all come from that fixed order.
Does provably fair change the RTP?
No. It is a verification layer over the published mathematics. The return to player and the volatility of a game are the same with or without it.
Can an operator switch models after launch?
Technically yes, but the model is part of how outcomes are produced, so a switch on a certified installation is a change a laboratory would need to review. Choose the model before launch and keep it.
The bottom line
Provably fair does not make a game more honest; a certified RNG already does that. What it does is make honesty checkable, which is a different and, in trust-sensitive markets, more valuable thing. Offering provably fair casino games in the model your players already recognise removes the largest single objection standing between a visitor and a deposit.
If you are deciding which provably fair casino games and which fairness model fit your market, the configuration wizard points you to the right package in a couple of minutes, or you can browse the full HTML5 catalogue and see the game types for yourself. Prefer to talk it through? Get in touch and we will walk you through how verification works on our platform.