Random Team Generator

How the Random Team Generator works

How the random team generator works

Splitting a group into teams sounds trivial and is surprisingly easy to get subtly wrong. There are two things to get right: the shuffle has to be fair, meaning every possible arrangement is equally likely, and the dealing has to be even, meaning no team ends up two people short because of how the numbers divided. This page does both, and the details matter enough to explain.

A fair shuffle. The correct algorithm is the Fisher–Yates shuffle: walk from the end of the list to the start, and at each position swap the current item with a randomly chosen item from the unshuffled portion including itself. Done that way, every one of the n! orderings has exactly the same probability. The common shortcut of sorting the list with a random comparison function does not do this; it produces some orderings far more often than others, and the bias depends on the browser's sort implementation. Team generators built that way tend to leave the first few names clustered together more than chance would predict, which is exactly the thing people complain about when they suspect a draw was fixed.

Where the randomness comes from. Each swap needs a random index, and there are two places to get one. Math.random() is a pseudo-random sequence from a seed, statistically fine but predictable in principle if someone can observe enough of its output. crypto.getRandomValues() draws from the operating system's cryptographic generator, seeded continuously from hardware noise and designed so that previous outputs reveal nothing about the next. This page uses the cryptographic source and converts its 32-bit output to an index with rejection sampling, discarding the small top range of values that would otherwise make low indices very slightly more likely. For a pub quiz the difference is academic; for anything where team assignment carries a prize or a burden, it means nobody can reasonably argue the draw was predictable.

Dealing evenly. Once shuffled, names are dealt round the teams like cards: first name to team 1, second to team 2, and round again. Ten people into three teams gives 4, 3, 3 rather than 4, 4, 2. When you ask for a number of people per team instead, the number of teams is the ceiling of people divided by size, so 11 people at 4 per team gives three teams of 4, 4, 3. The stats beneath the result show the team size range so an uneven split is obvious at a glance.

Input handling. Names are split on new lines or commas, trimmed, and blank entries dropped. Duplicates are kept, because two people really can share a name; if the list came from somewhere messy, run it through the duplicate remover first. Anything from two names up works, and asking for more teams than people simply gives you one person per team.

Reshuffling. Every change reshuffles from scratch, so if the first draw puts the two people who cannot be in the same room together, change the Reshuffle number for a fresh deal. That is a legitimate use; what it is not is a way to get a "better" random result, because there is no such thing. A fair shuffle will regularly produce teams that look lopsided by skill or friendship, and that is what random means.

Related tools