What provably fair proves:
Before you place your bet, the casino has already locked itself into a result. It generates a server seed, hashes it with SHA-256, and shows you the hash. That’s the commit-reveal protocol, a cryptographic commitment that means any post-bet alteration would produce a different hash and be immediately detectable. SHA-256 is specified in NIST FIPS 180-4, with 128-bit collision resistance. The computational effort required to find a collision makes tampering effectively impossible. This isn’t casino-proprietary cryptography; it’s the same standard used across the security industry.
The result itself is reproducible. Three inputs, server seed, client seed, and nonce, combine deterministically to produce the outcome. Anyone with those values can re-run the calculation and arrive at the same number. There’s no hidden variable.
Crash games take this further. Rather than committing to one outcome at a time, the casino generates a chain of millions of hashes recursively via SHA-256 and uses them in reverse order. The entire sequence of results is locked in before the first round plays. Each hash in the chain is verifiable by confirming that SHA-256 of hash N-1 produces hash N.
What provably fair does NOT prove:
That the outcome mapping formula matches the advertised RTP. This is the central gap. The random number generation is honest, verified, locked, and reproducible. But the formula converting that random number into a crash multiplier, a dice roll, or a card dealt? That’s a separate piece of logic, and it may or may not be published. A casino could operate a provably fair system with a 10% house edge while advertising 1%, and every individual verification would pass. The randomness isn’t the problem. The payout structure is.
That aggregate results match theoretical expectations. Provably fair verifies individual rounds, not distributions. Whether your results across a thousand rounds align with the claimed RTP is a statistical question, not a cryptographic one.
That the casino is trustworthy beyond game outcomes. Provably fair addresses outcome integrity only. It has nothing to say about withdrawal reliability, bonus term fairness, solvency, or data security. A casino with flawless provably fair implementation can still refuse payouts.
That provably fair replaces licensing. Licensing covers consumer protection, anti-money laundering, dispute resolution, responsible gambling protections, and fund segregation. Provably fair covers outcome verification. They’re complementary systems, not substitutes. A provably fair label doesn’t perform the functions of a Curaçao gaming licence or any other regulatory framework.
That your default client seed is safe. This one is actionable and often overlooked. If you never change the client seed and the casino generated it, the casino knows your client seed before committing the server seed. That means it could, in theory, select a server seed that produces a favourable-to-the-house outcome while still passing verification. The fix is simple: every major provably fair operator lets you set a custom client seed. Use it. Every session.
Layer 1: Traditional RNG + Lab Certification
The model governing the vast majority of online casino games. The casino runs a proprietary random number generator, and a third-party testing lab, eCOGRA, GLI, BMM Testlabs, or iTech Labs, audits the system periodically. The audits verify statistical randomness and confirm that payout distributions match the advertised RTP. It’s a closed system with full regulatory recognition across every major jurisdiction and decades of operational history. You can’t verify individual outcomes, your trust is anchored to the regulator and lab reputation. Between audits, you’re trusting that nothing changed.
Layer 2: Off-Chain Provably Fair
The commit-reveal protocol gives you per-round verification that the traditional model can’t match. You can prove, mathematically, that the casino didn’t change the outcome after you bet.
The history starts in April 2012, when SatoshiDice launched as the first blockchain-based betting game. It used Bitcoin transaction hashes as a primitive form of verification, ran a 1–1.9% house edge, and at its peak accounted for roughly half of all Bitcoin transactions. It wasn’t commit-reveal in the modern sense, but it established the principle that players could verify outcomes without trusting the house.
BitZino launched mid-2012 and was among the first to implement the commit-reveal model that became the foundation. By 2013 the method was formalised. PrimeDice and JustDice pushed adoption during the 2014–2016 golden era of Bitcoin dice, a game type ideal for provably fair because verification is trivially simple: one hash, one number. In February 2018, BGaming became the first major iGaming provider to introduce provably fair in online slots, implementing SHA-256 across a portfolio that now includes 35+ provably fair games.
The trade-off: the casino still controls the server. The outcome mapping formula may or may not be published. No major regulator formally recognises provably fair as a fairness standard. And there’s an IP protection tension that explains why many operators resist full transparency, a single game costs between $20,000 and $500,000 to develop, and publishing the complete source code risks cloning. ProvablyFair.org, an emerging independent certification body, is developing hybrid audit models that protect operator IP while enabling meaningful verification.
Layer 3: On-Chain Smart Contract Provably Fair
The next generation moves randomness fully on-chain. Instead of a casino-controlled server seed, randomness is generated by decentralised oracle networks, ORAO VRF on Solana, Chainlink VRF on Ethereum. No single party controls the outcome. The blockchain records everything.
Solana is emerging as a leading chain for this: sub-second finality, transaction costs under $0.001, and throughput of 65,000 transactions per second. Zero-knowledge proofs are also appearing, enabling verification without exposing proprietary game logic.
On-chain provably fair is where the technology is heading, but most implementations today are experimental. Latency and gas fees still constrain high-frequency games, the available game types are limited, and the ecosystem is small. The architecture is sound. The maturity isn’t there yet.
| Traditional RNG | Off-Chain Provably Fair | On-Chain Smart Contract | |
| Who controls randomness | Casino (proprietary RNG) | Casino (server seed) | No single party (decentralised oracle) |
| What the player verifies | Nothing directly | Individual outcomes via commit-reveal | All outcomes on-chain |
| What remains unverifiable | Individual outcomes between audits | Outcome mapping formula (unless operator publishes it) | Very little, code is public |
| Regulatory status | Fully recognised across all major jurisdictions | No formal regulatory recognition | No formal regulatory recognition |
| Game Type | Difficulty | Why |
| Dice | Trivial | One hash produces one number. A single formula, verifiable in 30 seconds. Stake’s dice formula: roll = (float × 10001) / 100 across 10,001 outcomes. |
| Crash | Moderate | Requires understanding of recursive hash chains. BC.Game’s crash formula takes the first 13 hex characters (2⁵² range), converts to a 0–1 float, applies 99/(1-X) with a 1% house edge. Verifiable, but you need to understand the chain mechanics. |
| Plinko | Moderate–Complex | Multiple derivations per game. Each ball drop requires a separate calculation. Tractable if the mapping is published, tedious if it isn’t. |
| Blackjack | Complex | Requires verifying a full deck shuffle via the Fisher-Yates algorithm. Stake simulates an unlimited deck with 52 equal-probability outcomes per draw. |
| Slots | Effectively impossible* | Reel strips, symbol weighting, and bonus trigger logic are usually proprietary. Without the outcome mapping, you can verify the seed was honest but not what result it produced. |
For most players, provably fair verification is practically meaningful for dice and crash. Everything above that on the complexity spectrum requires trusting the operator’s documentation, which is exactly the variable the next section examines.
| Operator | Outcome Mapping | Key Detail | Client Seed Control |
| Stake | Fully published | All game formulas and code published. House edge visible: 0.99 factor in crash = 1% edge, 0.98 in Drill = 2%. Uses HMAC-SHA256 (not HMAC-SHA512). Dice: 10,001 possible outcomes. Slots: reel lengths published. | Full. Player can set a custom seed; platform recommends it. |
| BC.Game | Partially published | Crash formula in white paper. Open-source GitHub verifiers. Other game types less documented. | Full. Player can customise. |
| BetFury | Not published | Verifier confirms seed compatibility only. No unhashed server seed exposed, no HMAC output, no formula. | Restricted, cannot change client seed. |
The difference between Stake and BetFury isn’t a matter of degree, it’s a difference in kind. When Stake publishes the code showing (2^32 / (int + 1)) * (1 – 0.01) for crash, you can read the 1% house edge directly from the formula. The verification closes the loop: you confirm the seed wasn’t tampered with and confirm the formula produces the payout distribution claimed. At BetFury, the verifier confirms the seeds are compatible, but without published formulas, without the ability to set your own client seed, and without visibility into the HMAC output, you’re verifying the lock without ever seeing what’s in the safe.
BC.Game sits between the two. The crash algorithm is well-documented in the white paper, the GitHub verifiers are open-source, and the verification for that specific game is solid. But the documentation for other game types is thinner, and the gap between crash documentation and everything else means the practical transparency depends on which game you’re playing.
When evaluating operators not on this list, the pattern to watch for: does the operator publish outcome mapping (the formula converting hashes into results) or only a seed verifier (a tool confirming the seeds match)? A seed verifier alone doesn’t tell you what result those seeds should have produced. It’s the difference between checking that the dice weren’t loaded and checking what number they landed on.
One finding from cross-platform analysis underlines the point: the same server seed, client seed, and nonce produce different results at Stake and BC.Game. The likely cause is different input ordering (Stake feeds the server seed first; BC.Game feeds the client seed first) or different HMAC implementations. Both systems are internally consistent, but “provably fair” isn’t a single standard. It’s a family of implementations, and the details vary operator to operator.
The UK Gambling Commission makes no mention of provably fair in any guidance document, consultation paper, or enforcement action. The UKGC also doesn’t permit cryptocurrency as a standard cashier option for UK-licensed operators, which puts most provably fair casinos outside its jurisdiction entirely. The Malta Gaming Authority has no specific guidance on provably fair either, BGaming holds an MGA B2B licence, but that covers broader operational compliance, not provably fair as a distinct standard.
Most provably fair casinos operate under Curaçao, Anjouan, or Kahnawake, jurisdictions where game testing requirements are minimal. This isn’t inherently a problem, provably fair is itself a form of transparency that traditional regulation doesn’t offer, but it means the two systems aren’t overlapping. You get one or the other, rarely both.
ProvablyFair.org is emerging as an independent certification body specifically for provably fair standards. It audits RNG and game logic and is developing hybrid audit models that protect operator intellectual property while enabling meaningful third-party verification. It’s the closest thing to a dedicated provably fair certification authority, though it operates outside the traditional licensing framework.
Provably fair and regulation address different problems. Provably fair covers outcome verification. Licensing covers consumer protection, anti-money laundering, dispute resolution, responsible gambling, and fund segregation. A provably fair casino without a licence is verified but unregulated. A licensed casino without provably fair is regulated but opaque on individual outcomes. The ideal is both. That combination is currently rare.
What to Check Before You Trust a Provably Fair Casino
Set your own client seed
If the casino generates your client seed and you never change it, it knows the seed before committing. That means it could select a server seed producing a house-favourable outcome that still passes verification. Every major operator lets you set a custom seed. If one doesn’t, as with BetFury, treat that as a transparency red flag.
Check for outcome mapping, not just a seed verifier
A verification tool that only confirms seed compatibility tells you the seeds were used correctly. It doesn’t tell you the formula converting those seeds into game results matches the advertised RTP. Look for the actual formulas, the code showing how a hash output becomes a dice roll, crash multiplier, or card dealt. If the operator publishes them, you can verify the house edge. If they don’t, you’re trusting their marketing.
Verify at least occasionally
Most people never check a receipt, but the option to check keeps the system honest. Pick one dice roll per session, re-run the HMAC calculation against the published formula, confirm the numbers match. An NSoft developer writing about provably fair implementation candidly noted uncertainty about whether players actually use verification, but the deterrent value is real. Operators who know verification is possible behave differently from operators who know it isn’t.
Understand the boundaries
Provably fair covers game-outcome integrity. It says nothing about whether the casino will process your withdrawal, honour its bonus terms, remain solvent, or protect your data. For those questions, you need to evaluate the operator’s licensing, track record, and payment security practices. See our review process for how we assess casinos across the full range of trust signals.