Farflung scoring starts with a correct answer
A versioned answer set determines whether a country matches a clue. Sources and scope belong to the question. A rare country does not earn points when it fails the clue, and a small country is not automatically rare.
Then, how unexpected is that answer?
Rarity is specific to a question. Editorial review supplies a baseline. When an AI assessment is configured and approved, it contributes 40% of the initial estimate; editorial judgment supplies 60%. Without an approved AI result, the game uses the editorial baseline.
Eligible results from earlier closed competitions can inform future scoring. Independent-player samples are smoothed, and their weight grows with sample count up to a maximum of 85%. Repeated practice guesses do not inflate those samples. Scores are frozen before a daily challenge opens.
The visible scoring label matters
Every result identifies its actual scoring method and sample count. “Editorial estimate” means that no approved AI assessment was used. “AI-assisted estimate” is still an estimate, not a measured live popularity percentage. A historical blend only appears when eligible past results exist.
No moving goalposts
Today’s players use the same question versions, order, rules, and scores. A later model response or another player’s answer cannot change your score. A verified factual error may void a round for everyone, with an audit record and a revised maximum score.
Ranking and corrections
Scores tie fairly. Speed is shown for your own review and never used as an undisclosed ranking advantage. See the leaderboard rules for weekly scoring and corrections.