Provably fair random picking
Most random pickers ask you to take their word for it. This page explains what we do instead, in the order it happens, and what each step would and would not stop. The normative version is the algorithm specification; this is the readable one.
Try it here
The tool below is set to this mode already.
Entries
More options
The three steps, in order
- Commit: your browser hashes the list, the weights and a random salt, and the server stores that hash before anything else happens. The list is now fixed, but still secret.
- Draw: the server asks PureRand for fresh random values. Each request is recorded in a public chain with a timestamp, so the randomness has an existence independent of us.
- Record: the commitment, the randomness, its source and the result go on a receipt at a permanent URL, which verifies in the reader's own browser.
What this does and does not prove
It proves that the list was fixed before the randomness arrived, and that the randomness came from a chain whose history we do not write. It does not prove that the list contained everyone it should have: if a name was never entered, no amount of cryptography will notice. That is why the publish option exists for draws where entrants need to see themselves on the list. Being clear about the boundary is part of the claim.
Questions people ask
- What exactly is a commitment?
- A SHA-256 hash of the entries, their weights and a random salt, in a canonical form. Publishing the hash fixes the list without revealing it, and the salt stops anyone from guessing a short list by trying candidates against the hash.
- Why does the order of the steps matter so much?
- Because it removes the two ways a draw is usually rigged. Committing first means the list cannot be rewritten once the result is known; drawing afterwards means the result cannot have been available when the list was chosen.
- Do I have to trust PureRand?
- Less than you would otherwise. The randomness is recorded in a chain with timestamps and hashes, so a value cannot be quietly replaced afterwards. What the chain does not do is prove a value was unpredictable at the moment it was produced; that rests on the source, and the receipt names it.
- Can I verify a receipt without this site?
- Yes. The specification includes a reference implementation in Python and a set of test vectors; the receipt has everything it needs as input. The verifier on this site runs the same computation in the browser.
What the receipt proves
Every draw is committed before it happens and recorded afterwards. The receipt holds the commitment, the randomness, its source and the result, and anyone with the link can check that they fit together.