Random Order Shuffler

A Fisher-Yates shuffle that respects whatever separator your list already uses — newlines, commas, semicolons — so a column pasted from a spreadsheet comes back shuffled in the same shape it went in. Useful for assigning code-review pairs, randomising demo order, drawing a draft order, or anything that needs a fair permutation rather than a vague sense of mixing.

Leave the seed empty and each shuffle draws fresh bits from crypto.getRandomValues, the operating system's cryptographic generator. Fill the seed in and the tool switches to mulberry32, a small deterministic generator, so the same seed and the same list always produce the same order. That is the useful half of a seeded draw: you can publish the seed and let anyone reproduce the result instead of asking them to take your word for it.

The reproducibility is only worth something if the seed is committed before the draw. A seed chosen after seeing the list can be searched — try a few thousand until one puts the right name first, and publish that one. The honest procedure is to announce the seed, or better a rule that fixes it to something nobody controls (a closing price, a lottery number, a block hash), and only then run the shuffle. Published afterwards, a seed proves the arithmetic and nothing about the intent.

There is also a size limit on what a seeded shuffle can reach. Mulberry32 holds 32 bits of state, so there are about 4.3 billion distinct seeds — which is fewer than the number of orderings of thirteen items, since 13 factorial is 6.2 billion. For a list of twelve or fewer, every possible order is reachable; past that, most orders simply cannot be produced by any seed. It does not make a particular draw unfair, and nobody can exploit it without an enormous search, but it means "every arrangement is possible" stops being literally true somewhere around a baker's dozen. The unseeded mode has no such ceiling.