Dealing rather than drawing

There are two ways to split a list into groups and they behave very differently. Shuffling the list once and dealing it out round-robin gives groups whose sizes never differ by more than one. Choosing a random person for each slot independently gives groups that are frequently lopsided — and with six jobs and four people, independent choice regularly leaves one person with three and another with none.

Dealing is the version people mean when they say random, and it is not what an independent draw produces.

Why does independent choice go wrong?

Because each draw knows nothing about the ones before it. It is sampling with replacement, and the coverage problem that follows is the same one the picker article describes for names.

Dealing over a shuffled list is sampling without replacement, and it enforces evenness by construction rather than by luck. Eleven people into three teams gives four, four and three — never five, five and one.

The distinction is invisible in a single use and obvious over a term, which is why a rota built on independent draws generates complaints and one built on dealing does not.

Which mode do you actually want?

The two options answer different questions and picking the wrong one is the commonest source of an awkward result.

For classroom work, fixing the size is almost always right: the activity was designed for groups of three or four, and the number of groups is whatever the register produces. For a league, the number of teams is the constraint and the sizes fall out of it.

What happens with an odd number?

It depends on what a leftover person costs. For pairs, the last group becomes three rather than leaving someone without a partner — which is what you want in a classroom and would be wrong in a chess tournament.

For teams, the remainder is spread so no team is more than one person short. For a bracket, byes are added to reach the next power of two: eleven entrants become sixteen slots, so five competitors get a first-round bye.

Each of those is a policy choice rather than an arithmetic one, and stating it in advance is what stops it being read as favouritism after the draw.

Why is Secret Santa not just a shuffle?

Because a naive draw can produce mutual pairs — Ana and Ben buy for each other while the rest of the group is unaffected — and it can hand someone their own name.

The fix is to form a single chain: Ana gives to Ben, Ben gives to Chiara, and the last person gives back to Ana. Everyone gives once and receives once, nobody draws themselves, and there are no closed pairs.

That is a cyclic permutation rather than a shuffle, and it is the reason a dedicated generator produces a better draw than pulling names out of a hat with a redraw rule bolted on.

Constraints are the thing none of these tools model, and they matter more than the randomness does. Two people who should not be paired, a group that needs one confident speaker, an entrant seeded to avoid a first-round repeat — all of them are real requirements that a uniform draw will violate roughly as often as chance dictates.

The workable pattern is to draw and then adjust visibly, rather than to redraw until the result happens to satisfy an unstated rule. A visible adjustment is a decision anyone can question; a hidden redraw is a draw nobody can check.

What makes a shuffle actually uniform?

Fisher–Yates, which walks the list once swapping each element with a randomly chosen one at or before it. Every ordering is equally likely and it runs in a single pass.

The obvious alternative — assigning each item a random key and sorting — sounds equivalent and is not uniform in general, particularly with a comparison sort and duplicate keys. It is one of the classic subtle bugs in this area, and it produces distributions that look fine until someone measures them.

Both matter more than they seem for anything that repeats. A biased shuffle used once is invisible; used every week for a year it produces a pattern people will notice and cannot explain.

Questions people ask

Should seating be random? As a blunt instrument, yes — it breaks up the habits that form in the first week without anyone being singled out. It is not a substitute for deliberate seating where there is a specific reason for one.

How are byes assigned in a bracket? The number of byes is the next power of two minus the entrant count, and they are distributed so that no two byes meet in round two — otherwise a competitor advances twice without playing.

Is round-robin dealing fair over time? Within one deal, yes. Across repeated deals the same person can land in the same slot twice by chance, which is why resetting and re-dealing each round is different from rotating a fixed order.

Can I weight a draw? Not in these tools, and duplicating an entry is the transparent way to do it — the weighting is then visible in the list rather than hidden in the algorithm.

Shuffle once, deal round-robin, and state the leftover rule before you draw. The team, group and pair generators cover the two fixing modes, Secret Santa builds the single chain, and the order, bracket, seating and task assigner tools apply the same shuffle to the other shapes the question takes.