How we score a quiz

No model reads your answers

A language model would give a different reading of the same answers every time, and nobody could check it. Our scoring is arithmetic you could redo by hand from the weights.

We do use AI to draft questions. An editor rewrites and approves every line before it publishes, and the scoring never involves a model.

What has to be true before a quiz publishes

Every quiz runs twenty thousand simulated answer paths. It does not ship unless all of these hold:

• Every outcome is reachable. Each of the four results has real answer paths that produce it.
• No outcome dominates. None may win more than 55 percent of simulated paths, and none may fall below 8 percent.
• At least one outcome is genuinely uncommon. On a four-outcome quiz, one result lands at 15 percent or less. Without that, rarity means nothing.
• Every question changes results. If switching your answer never changes anyone’s outcome, the question gets cut.
• Ties break on your answers. When the top two finish within two points, the earliest answer that separates them decides it, not alphabetical order.

Where the percentages come from

A result page shows how common your result is. Until a quiz has 200 finished attempts, that number comes from the simulation and is worded as a fact about the quiz, not about people. Past 200 attempts, the live count replaces it and the page says how many attempts it was counted from. We never invent a statistic, which rules out some popular headlines.

What this is not

These are entertainment. They are not psychological assessments and they measure nothing clinical. We do not infer anything about your health, sexuality, race, religion or politics, and we do not build quizzes that try.

Corrections

If a result seems wrong, tell us which quiz and what you expected: hello@quizfound.com.