The Lab · interactive

Provably fair, verified: what the hash check proves and what it does not

Predict what a 'provably fair' label proves, then run a commit-reveal round with real SHA-256 in your browser. You will see the check catch a swapped seed, and see that a round which passes it still carries a house edge.

loading experiment…

How it works

Each round follows the commit-reveal pattern. The server draws a secret seed of 16 random bytes and shows only its SHA-256 hash, the commitment. You add a client seed, generated at random unless you type your own, and a nonce, fixed at 1 because every round here gets its own server seed. The outcome is the SHA-256 hash of 'serverSeed:clientSeed:nonce': its first 52 bits become a fraction r between 0 and 1, and the crash multiplier is the whole-number part of 99 ÷ (1 − r), divided by 100, with a minimum of 1.00×. Changing the client seed recomputes the outcome but keeps the committed server seed, since a server allowed to draw a new seed after seeing yours could pick one that suits it.

After the prediction, the tool shows the commitment, the client seed and the resulting multiplier. 'Reveal the server seed' discloses the committed seed; 'Reveal a swapped seed (a cheating server)' discloses a different, freshly drawn one instead. 'Re-hash & check' hashes the revealed seed, compares it with the commitment shown earlier, and re-derives the multiplier from the revealed seed, the client seed and the nonce. The honest seed passes both tests. The swapped seed fails, because its hash is not the commitment: NIST's Secure Hash Standard calls SHA-256 secure because finding a message for a given digest, or two messages with the same digest, is computationally infeasible. 'New round' deals a fresh committed seed.

A passing check proves integrity: the outcome follows from inputs fixed before the round, through a disclosed algorithm, and was not altered afterwards. It does not prove a fair price. Under this formula the multiplier reaches at least m× with probability 0.99 ÷ m, so a 2× target is reached in 49.5% of rounds and a bet cashed out there returns 2 × 0.495 = 0.99 of the stake on average; the same holds for any target above 1.00× in the 0.01 steps the multiplier moves in. That is a 1% house edge, built into the maths and fully verifiable. Nor does the check show how the server seed was generated: GLI-19, a test-laboratory standard, requires it to come from a cryptographically strong RNG, and notes that verifying a provably fair outcome does not by itself establish statistical randomness, unbiased outcome distributions or compliance with the approved game mathematics. It also says nothing about whether an operator is licensed.

What it leaves out: Each round has its own server seed and a nonce of 1, so it leaves out seed rotation, in which one committed seed covers many rounds and its reveal may wait until it is retired or rotated, as GLI-19 permits. There is no stake, cash-out or rising multiplier, and the server and the verifier run in the same browser, so it shows the logic of the check rather than auditing any operator's implementation. Meant for adults (18+) as an educational model, it cannot change the house edge of any real game.

Sources

Checked on 2026-11-09. Independent, not affiliated. 18+.

All Lab experiments →