Provably Fair Games: What They Are (and Why Most Crypto Games Aren't)

Strip the marketing and the provably fair meaning is narrow and specific. It means the game commits to a hidden value before you play, then reveals it afterward, so you (not a regulator, not an auditor, you personally) can recompute the outcome and check nothing was swapped mid-round.

Compare that to a normal online casino or a normal game backend. There, the house rolls the dice on a server you can't see, tells you the result, and you either trust it or you don't. Provably fair replaces "trust us" with "check it." That's the entire pitch. Whether a specific game delivers on it is a separate question, and honestly, a lot of them don't.

The term shows up most in crypto gambling (dice, crash, roulette), but the mechanism is the same one that would let a card game or a PvP arena prove its results too. Same cryptography, wildly different scope, which is exactly where the trouble starts.

Here's how the standard version works, minus the jargon soup.

A commit-and-reveal scheme is a two-phase protocol. First the operator commits to a secret by publishing its hash. Then, later, it reveals the secret, and anyone can hash that secret again and confirm it matches the commitment shown earlier. Two properties make this work: the hash leaks nothing about the secret beforehand (hiding), and the operator can't quietly swap the secret afterward without the mismatch being obvious (binding).

In a casino-style game, a few inputs go into each round. The server seed is the operator's secret, and its hash gets shown to you before you bet, so it's locked in early. The client seed comes from you, or from your browser, so the operator can't precompute an outcome tailored to you specifically. Then there's the nonce, a counter that ticks up with every bet, so the same seed pair still produces a different result each time.

As Chainlink's own technical guide lays out, these systems usually feed the server seed, client seed, and nonce through HMAC-SHA256 to produce the number that decides the round. After you're done, the server seed gets revealed, you re-run the same function on the same inputs, and you confirm the result. If it matches, the round was honest. If it doesn't, someone cheated and the math rats them out.

On-chain games sometimes add a verifiable random function on top. A VRF produces a random output plus a cryptographic proof anyone can check, without needing to trust whoever generated the number. It's the same "don't trust, verify" idea, wearing a smart-contract coat. That combination, a random output plus a checkable proof, is what people mean by a verifiable RNG. Want to actually run one of these checks on a real result rather than read about it? That's its own walkthrough, and we've got a step-by-step on using a provably fair checker for exactly that.

One caution before we move on. The badge is not the proof. Some games publish the full formula and every input so you genuinely could recompute a round. Others slap "provably fair" on the marketing page and never publish an independently runnable algorithm at all. If there's no public formula, there's nothing to verify, and the label is just vibes. That alone is a decent rug red flag worth watching for.

This is where most people get it backwards, so read this part twice.

Provably fair proves that the process wasn't altered after you committed to it. That's it. That's the whole guarantee. It says nothing about whether the odds are in your favor.

Chainlink is blunt about this: provably fair randomness guarantees the outcome wasn't manipulated, but it does not guarantee you'll win. A game can be fully provably fair and still carry a brutal house edge. The dice weren't loaded. The dice were just always going to beat you over time, out in the open, honestly. Verifiable just means checkable. It says nothing about whether the odds favor you.

It also doesn't mean uncheatable. One of the earliest academic papers on the subject points out that provably fair smart-contract casinos can still be gamed by attackers with enough money and computing power. Provable and unbeatable are different claims. Provable means checkable, not invincible.

And "on-chain" isn't a synonym for fair either. Blockchains can't natively produce secure randomness, so most of the real randomness happens off-chain and only the proof touches the chain. Worse, as Fhenix's technical blog argues, on-chain visibility can actually work against fairness, because publicly visible player actions let bots front-run and exploit them in real time. A verified random number says nothing about that attack surface. Transparency in one place doesn't buy you fairness everywhere.

Quick detour, because this trips up smart people constantly.

Publishing the odds and letting you verify a result are two completely different promises. Apple's App Store guidelines have required loot-box odds disclosure since December 2017. Google, China, and South Korea have their own versions. But disclosure only means a probability gets printed somewhere. It gives you zero way to confirm that your specific pull actually matched that number.

South Korea even enforces its version, and a 2025 compliance study found that 84.4% of surveyed South Korean mobile games with loot boxes disclosed their odds as legally required. Good, sort of. Except disclosure still never extends to letting a player verify a single pull against the stated number. You're told the odds. You're trusting that the machine actually used them. That's the honor system with a legal wrapper, not provable fairness.

So when a game tells you "our drop rates are published," nod politely, and then ask the harder question: can I check that this particular result matched them? Usually the answer is no.

Now the part the entire ranking SERP somehow misses.

Almost every "provably fair" crypto game applies the commit-reveal discipline to one narrow thing: the loot roll, the pack open, the single dice throw. And there's real activity behind that layer, it's not vaporware. A peer-reviewed PLOS ONE study tracked over 2.2 million transactions across three decentralised gambling contracts, including the provably-fair dice game Etheroll. Provably fair dice have years of on-chain history you can go read.

But a card battle isn't a dice throw. A PvP match is a long sequence of RNG rolls, ordering decisions, hidden state, and rules interactions stacked on top of each other. Proving that one number inside that sequence was honest doesn't prove the whole match resolved the way it should have. That's the sleight of hand. The badge covers the pack. The match, the thing you're actually competing over, mostly runs on trust-us server logic you can't see.

Is there a mechanism that could fix this? Yeah, and it already exists outside crypto. In deterministic games like Mechabellum, players have shown that the replay stores only the players' decisions plus the prerolled seeds, and the client recomputes the entire battle fresh from those inputs. Same seed, same recorded decisions, byte-identical outcome, every time. That property, an entire match being independently recomputable rather than just one roll inside it, is the actual bar. It's rare because it's hard, not because it's impossible.

This is also where DegenLoot's companion piece #502, "Provably Fair Gaming in 2026: What It Proves (and Doesn't)," goes deeper on what a given claim really certifies in practice. And if you want the shortlist of games that actually extend verifiability past the loot layer, we ranked them by what you can genuinely verify so you don't have to take anyone's badge at face value.

So does anyone apply the full commit-reveal discipline to the match itself and not just the pack? A few projects are trying. Stellarch (stellarch.io) is one worth a look, because its fairness axis is the exact gap this article is about.

Stellarch is a deterministic, seeded-RNG browser trading card game on Hive. You draft a 7-card team (6 Fighters plus a commander) and battle, and the combat resolves entirely through a seeded-RNG engine. That means every match replays byte-identically from its seed. Same teams, same mana cap, same ruleset, same winner, every single time. You verify the result from the seed instead of trusting the operator's server. That's the property most "provably fair" games stop short of, applied to the whole match rather than one roll.

A few things that make it easy to poke at yourself, honestly rather than as a pitch:

• It's free to start. A new account mints no cards and needs no purchase, because the draft pool is a ghost-catalog floor you can build a full legal team from right away.

• No wallet required to try it. You can sign up with email or Google, or use Hive Keychain if you want. See our roundup of crypto games you can try without a wallet for more that skip the gate.

• It runs in the browser. Nothing to download.

• It's on Hive, the same chain family as Splinterlands. There's real depth already, 290+ Fighter and commander cards across 8 affinities, not a tech demo.

Now the honest boundary, because this is DegenLoot and we don't do rigged verdicts. Stellarch is in closed alpha with a waitlist right now, not a full public launch. Crypto deposit and withdrawal rails are off, so nobody's cashing out today, and ranked token earning isn't switched on either. If you came looking for a play-to-earn payday, this isn't that, and pretending otherwise would be exactly the kind of dishonest hype this whole article is arguing against.

And credit where it's due: Splinterlands wins on player base and market depth, full stop. It's the bigger, more established browser TCG, and that's not close. See how it stacks up against other browser card games if size and ecosystem matter more to you than the fairness axis. Stellarch's honest edge is seed-verifiable match outcomes and a genuine free start, not size. If that specific thing interests you, join the waitlist for early access and go verify a replay yourself. That's the only sales pitch that fits a post about checking the math instead of trusting the badge.

Because that's the takeaway underneath all of it. Provably fair games are a real, useful property, but a narrow one, and most only ever apply it to the easy half. Learn what the words actually promise, then check whether a game publishes something you can run. The badge is the start of the question, not the answer.