Every fairness argument in a pickup game or a classroom starts with a feeling: "they always end up together." So we stopped arguing and counted. Ten thousand seeded splits of the same 24-player roster, every one reproducible from its seed, tracking a single pair through all of them — here is what the numbers actually say.
Every figure in this post is a build-time output of the same random team generator engine — the same seeded shuffle, the same round-robin deal, the same history-aware re-roll logic. The roster is 24 fixed names split into 4 teams of 6, and the tracked pair is Ava and Liam. Each of the 10,000 draws used its own deterministic seed, so the whole simulation is reproducible line for line.
The three numbers that settle the argument
| Question | Analytic answer | Simulated (10,000 draws) |
|---|---|---|
| Do Ava and Liam share a team in a single draw? | 5 ÷ 23 = 21.74% | 2,163 / 10,000 = 21.63% |
| Do they share a team in two independent draws in a row? | 4.73% | 2.90% |
| Same question, with history-aware redraw on the second draw | engine target: 0 | 9999 repeats in 9,999 draws = 0.00% |
The analytic column is not an approximation of the simulation — it is the exact value, derived from seat-counting: fix Ava, and 5 of the 23 remaining seats sit on her team. The simulation agreeing to within a fifth of a percentage point is simply a check that the engine implements the math it claims to.
Why "they always end up together" feels true
One draw in five is not rare. In a single draw, any given pair shares a team 21.74% of the time — close to a coin flip happening twice. A class of 24 contains 276 distinct pairs, and every one of those pairs is someone's "they're always together." Expect several same-team pairs in every single draw; the surprising outcome would be a draw with none.
Streaks cluster more than intuition expects. With independent redraws, the same pair collides in consecutive draws 4.7% of the time — about twice per school year at two practices a week. When it happens to the same visibly annoyed kid two weeks running, it feels rigged. It isn't; it's the square of one-fifth arriving on schedule. The fairness methodology page covers why verifiable seeds, not vibes, are the fix.
The history knob has a real cost, and a real receipt. Passing the previous draw in as history drove the measured back-to-back repeat rate from 4.7% to 0.0% across all 9,999 consecutive draws, costing an average of 8.6 internal re-rolls per draw. Perfect pair-avoidance is no longer raw randomness — the space of possible splits is quietly constrained — which is precisely why the seed stays on screen: anyone can verify that the avoidance happened, instead of trusting it.
Balance is a guarantee, not a tendency
Across all 10,000 draws, the largest team-size gap ever observed was 0. Round-robin dealing makes evenness structural: sizes can differ by at most one whenever the roster doesn't divide evenly, and by zero when it does. If you've ever played a game that came out 9-vs-4, the tool that produced it wasn't shuffling — it was counting off, which is deterministic and anything but random.
Run your own version in 30 seconds
- Open the random team generator and paste any roster you like.
- Draw once, note the seed, and pick any pair in the results to track.
- Redraw a few times — then re-run an earlier draw from its seed to confirm the same teams come back.
That's the whole trust model: the arithmetic above is public, the algorithm is published, and every draw carries the evidence needed to check it.