A judge that can grade free answers can also score things a point-per-answer game never could: how rare the knowledge was, how sure you were, how close you got, and how far you beat your own odds. Each rule below is chosen so that the best strategy is simply to play honestly.
p is the share of players who get the question right.
Why it’s fair. A question everyone knows shouldn't be worth the same as one almost nobody knows. This is the same bits-of-surprise rule Categories+ uses for rare answers.
q is the chance you gave to what actually happened: your confidence if you were right, one minus it if you were wrong.
Why it’s fair. It's a proper scoring rule: your expected score is highest when you report exactly how sure you are. Claiming more certainty than you have loses points on average.
Every doubling you're off costs two points.
The answer is 193. The slider moves on a log scale, like the scoring.
Why it’s fair. Twice too high and half too low cost the same, which matches how people think about size. Units are converted before scoring.
θ is your rating, b the question's difficulty, a how sharply it separates players. Result is 1 if right, 0 if wrong.
Why it’s fair. Every player's expected score is zero, so a kid and a professor at one table are in a fair fight. Golf calls the same idea strokes gained.
k is how many of the N players got it right.
Why it’s fair. Trivial questions and impossible ones both score zero. The only way to win is a fair question that splits the room.
y and n are the yes and no shares sold so far; b sets how fast prices move. A trade costs the change in C.
You hold 0 yes, 0 no. Prices are in chips per share; b = 10 here, for example.
Why it’s fair. This is a logarithmic market maker: it always takes the other side, and prices move only when someone backs a belief with chips. The answerer can't trade on their own question.
The model never supplies an answer from memory, and anything that can change carries an as-of date.
Every ruling cites the key version. A winning challenge becomes a proposed alias for an editor to approve.
Questions that players of equal skill answer very differently by age or region are flagged and kept out of mixed-age ranked play.
Once a question has been seen by enough tables, it leaves ranked play so memorizing the bank doesn't win a league.
Zebra can't see phones, so league nights lean on estimation, connections, show-your-work and speed rounds, which are hard to look up in time.
Calibration uses right-or-wrong counts per question. No audio or transcripts leave the device.