Probability theory / Applied overthinking
A rigorous theory of
very bad choices.
On the statistical consequences of following the crowd.
Abstract
We present a remarkably elaborate explanation of a remarkably small idea: every answer gets one ticket, and every player who picks it adds another. Draw a ticket. That answer loses. The rest of this paper exists because we found the equation font.
Keywords: probability, crowd psychology, suspicious confidence, KFC.
Axioms, or “the rules.”
Let there be a room, a set of answers, and several people who are certain they can outsmart a random number.
- N
- The number of answers available this round.
- ni
- The number of players who pick answer i.
- M
- The total number of completed picks, so M = ∑ ni.
- wi
- Answer i’s weight: one base ticket + its picks.
- L
- The randomly drawn losing answer. A deeply unpopular letter.
All base tickets are equal. Human and bot picks each add one ticket. Once everyone locks in or times out, the server draws uniformly from the tickets. An unchosen answer keeps its base ticket.
The crowd-stacking theorem.
Theorem 1. A more popular answer is more likely to lose. No answer is completely safe.
Proof. There are N base tickets and M player tickets. Answer i owns 1 + ni of them. Dividing its tickets by all tickets gives equation (1). This is the part where we adjust our glasses.
All probabilities add to 100%. More picks mean more tickets; equal picks mean equal odds. If no one picks an answer, its chance is still 1 / (N + M), which is greater than zero.
Corollary 1.1. “Nobody would pick that” is a strategy. “That can’t possibly lose” is a mistake.
If someone joins your answer, both its weight and the total increase. The change in your risk is strictly positive for a pool of at least two answers:
Here W = N + M. A pick on a different answer increases only the denominator, so your risk goes down. Hence the moving odds after you lock in.
Q.E.D. We divided two numbers, but in Latin.
Experimental validation.
A controlled study in three restaurants and poor judgment.
Submit your peer review.
Move the crowd. Watch the “advanced mathematics” happen.
Percentages are rounded to one decimal place; the total uses unrounded probabilities.
The result goes here. Your hypothesis is safe. For now.
Fig. 1 — The same probability and weighted-draw functions used by the game. This sandbox lets you see every pick; an actual round keeps other players’ choices hidden until everyone locks.
The progressive dread principle.
The answer pool follows 100 → 50 → 25 → 10 → 5 → 3 → 2. Holding the crowd fixed, fewer answers mean fewer base tickets and a larger chance that your answer loses.
| Answers | Only you pick it | All six pick it |
|---|---|---|
| 100 | 1.9% | 6.6% |
| 50 | 3.6% | 12.5% |
| 25 | 6.5% | 22.6% |
| 10 | 12.5% | 43.8% |
| 5 | 18.2% | 63.6% |
| 3 | 22.2% | 77.8% |
| 2 | 25% | 87.5% |
The other five players choose other answers in the middle column. The last row still has six players, so it is not a final duel. These are per-answer risks, not a promise of elimination each round.
Player counts also change as people drop out, so your personal odds need not rise every round. The table isolates the effect of shrinking the pool. We prefer our drama mathematically honest.
Necessary complications.
Information is not evenly distributed. So we hide it.
Before locking, you see a possible risk range. After locking, only your answer’s provisional risk and pick count update. You cannot switch. Once everyone is locked, the full distribution appears; the losing answer is drawn and its final percentage is shown at reveal.
Nobody loses? Shrink. Everybody loses? Replay.
A losing answer can be unchosen. Everyone advances. If every remaining player picks a losing answer, everyone is restored and the same-sized round repeats with new picks and a fresh draw.
Two players ≠ an immediate duel.
Two survivors keep choosing freely at five or three options. Only at the two-option stage does each receive a different answer. Equal weights then give each player exactly ½ risk, and exactly one wins.
No duplicates: exclusive claims, familiar arithmetic.
Each answer can be claimed once. Occupied answers have two tickets, empty ones have one. The pool stays large enough for every survivor to claim a legal answer.
Appendix A: Chaos, for readers who wanted more symbols.
Chaos draws two distinct losing answers without replacement. An answer can lose on the first draw, or survive that draw and lose on the second. With W total tickets, those mutually exclusive routes give:
For weights 3, 2, and 1, the first answer has an 85% chance of being one of the losers. The individual probabilities sum to 200% because two distinct answers are selected. Chaos keeps at least three options until the two-player, two-option final duel, which uses one loser.
Outguess the crowd.
Then blame the math.
You cannot guarantee safety. You can avoid the pile-on.
The rest is a random draw with excellent presentation.