The scoring method has one purpose: a player should be able to point at any part of a total and name the fact that produced it. The result panel therefore shows every pair, not only the pairs that earned points. This page publishes the same rules used by the game engine, so a hand calculation and a browser total can be compared without guessing.
Trait rarity becomes points
For every pair of cards, the game finds the traits they share. Each shared trait receives a weight based on its frequency in the library:
w = 10 × (1 − freq / N)
freq is the number of cards in the library carrying that trait, and N is the number of cards in the whole library. A trait found on 40 of 400 cards produces 10 × (1 − 40/400) = 9. A rare trait found on only 8 cards produces 9.8. The weight is attached to the trait, not to the pair, so the same fact has the same value whenever it appears.
The trait families are fixed. Field, nationality, era, medium, award, and role traits can establish a hand type. Place and institution traits, along with single work names, are deliberately weaker: they still contribute points to a pair, but they cannot by themselves build a pair, three of a kind, or any higher matching pattern.
How a hand adds up
The base value comes from the best hand type the five cards satisfy. The nine values are 0, 10, 25, 40, 60, 80, 100, 130, and 200. Only one base value is awarded. If a hand fits both a flush and a pair, the flush base applies because the higher pattern wins.
Each of the ten possible pairs then contributes half of the sum of its shared-trait weights. If a pair shares two traits with weights 9 and 7, that pair contributes 0.5 × (9 + 7) = 8. A pair with no shared trait contributes zero. The full hand is the base plus all ten pair contributions.
This is why the result panel has ten rows. Four of five cards might be related while the fifth is an isolated distractor. The scorer does not throw that fifth card away. It still runs all ten comparisons and shows which ones failed.
Follow the worked example
Consider five actor cards from an illustrative library of 400. The first pair shares field:actor with frequency 40, so its weight is 9 and its contribution is 4.5. The third and fourth cards share place:new-york with frequency 120, so its weight is 7 and its contribution is 3.5. No other pair shares a trait.
All five cards have the actor category, so the hand type is a flush with base 80. The total is 80 + 4.5 + 3.5 = 88. The page repeats this example instead of hiding the arithmetic in a single result.
Why not just count shared traits
Counting every shared trait equally would make common facts dominate. Two broadly popular categories could produce a large score while a genuinely unusual connection counted for very little. Rarity weighting reverses that incentive: the unusual shared trait is worth more, while a trait found almost everywhere is worth less.
The pair structure adds a second check. A hand must build a recognizable shape, not merely collect isolated coincidences. Five cards with one shared, common trait can still score, but the hand-type ladder rewards structure separately.
Why a hand can score zero
Zero means no pair in the five-card hand shares even one whitelist trait. The scorer still returns ten rows, all with zero. The explanation then identifies the closest pair and a trait that one card has but the other lacks. This turns an empty score into a specific observation about the hand.
The zero state is not an error. It is the bottom of the published scale, and it stays visible with the same pair table as every other result.
The trait and round caps
No single trait can contribute more than 30 points before the pair's half multiplier. The cap is a hard stop, not a multiplier. The complete hand is then capped at 999 points. Both limits are applied by the same scoring function used for the exhaustive best-hand check, so a malformed or inflated total fails validation rather than reaching the score board.
Version history
The current formula is score-v1, effective 2026-11-18. Every puzzle record stores its own rule_version and library_version. A replay is scored with those stored versions; a newer formula is not allowed to rewrite an older result. The changelog above is intentionally short because this is the first published version. Future changes must add a new version entry rather than silently changing an old one.