Splitting a department into project squads by hand takes a meeting; doing it randomly takes a click — and removes the "why am I always with the same three people?" conversation entirely. The demo below is a 12-person team with seniority weights (5 = principal, 1 = intern), pre-set to deal 3 balanced squads. Swap in your own people.
12 names entered · lines without a weight count as weight 3
Snake draft on skill weights ("Name, 3" = weight 1–5).
Everything runs in your browser — the names you type are never sent to a server. The share link encodes the roster in the URL itself, so only share it with people who may see the list.
The demo department has 35 seniority points across 12 people: two 5s, two 4s, three 3s, three 2s and two 1s. 35 does not divide by 3, so a perfectly equal split is impossible — the best any assignment can do is 12/12/11, and the snake draft lands exactly there. With this page's build-time seed it produced Squad 1: Ines, Mei, Hana, Zoe (12 pts); Squad 2: Raj, Drew, Luca, Kai (11 pts); Squad 3: Sofia, Tom, Jess, Omar (12 pts). Because round one hands the top three picks to three different squads, the two principals (the 5s) can never share a squad — that separation is structural, not luck. A pure random deal on this roster can, and regularly does, put both 5s together.
Computed by the same size logic the tool uses — sizes never differ by more than one:
| Groups | Sizes | Typical use |
|---|---|---|
| 2 | 6, 6 | Debate formats, build-vs-review pairs of squads |
| 3 | 4, 4, 4 | Hackathon squads, parallel workstreams |
| 4 | 3, 3, 3, 3 | Workshop tables, design-sprint pods |
| 6 | 2, 2, 2, 2, 2, 2 | Pair programming, peer-review pairs |
The same engine drives the random team generator (all modes, no preset), the 5-a-side team picker for the office football league — same snake draft, sport-flavored — and the Secret Santa generator for December, which adds a no-self-draw guarantee and private reveals. Running a training session? The classroom group generator has one-tap pair and group-of-4 presets that work just as well for adults.
Random for anything social or low-stakes: lunch roulette, retro breakouts, brainstorm pods — speed and visible neutrality matter more than composition. Balanced when output quality depends on the mix: hackathon squads, on-call rotations, workshop tables that each need a senior voice. Captains when leads are already decided — each project owner goes on one of the first lines and gets a randomly dealt squad, so there is no drafting meeting.
Weights are just numbers after names in a textarea — nothing forces you to project the input. Prepare the roster privately, generate, and share only the resulting groups (Copy teams strips nothing, so delete the "(n)" weight markers from the pasted text, or screenshot the cards before toggling balanced mode off). Also remember the share link contains the full roster including weights — share results, not the link, if scores are sensitive.
There is no exclusion engine — deliberately, because a half-reliable one is worse than none. The practical pattern for one or two constraints: generate, and if a forbidden pairing appears, swap one of those people with any same-size group member. One swap preserves every group size, and doing it manually keeps the rule visible to you instead of buried in settings.
The table on this page shows all even options. The trade-off: 4 groups of 3 maximizes airtime per person (good for discussion), while 3 groups of 4 tolerates a no-show better — losing one person from a trio changes its dynamics more than losing one from a quad. For pure pair work, 6 pairs, and re-draw between exercises so pairs rotate.
Use judgment. Weights here are a grouping convenience, not an assessment record — but a leaked roster reading "Kai, 1" reads like one. For anything beyond casual balancing, use neutral proxies instead: years in the team, timezone, or simply which product area someone knows. The algorithm neither knows nor cares what the number means.