A strong bridge reaches every clue.
Your answer is compared with each clue separately. A word that connects well with both clues can outscore one that fits a single clue brilliantly. In a three-word bridge, all three connections matter.
These illustrate connection strengths, not rankings among players.
There can be more than one full-score answer.
Every round has a featured connection and an explanation for each clue. We also recognize selected alternatives: down connects a bird and a pillow, for example. Recognized answers receive 1,000 points. An alternative bridge with its own explanation shows that route alongside the featured one.
Common inflections and selected British and American spellings are recognized. “Leaves” can express the same connection as “leaf”; “colour” and “color” are treated alike. Repeating a clue or its inflected form does not count as a new answer.
The final round gives you a meaning to work with.
A word can have several meanings. When the cue asks about intelligence, bright points toward clever or brilliant. Shiny belongs to another sense. Read the short cue above the answer box before choosing a word.
These rounds use reviewed alternatives and a specific dictionary sense. Known opposites receive at most 100 points. For other answers, the estimate considers both the clue and the selected alternatives, which helps reduce matches to the wrong meaning.
What happens to an answer we have not listed?
It still receives a semantic estimate. A local model compares patterns of word use, adjusts for each clue, and combines the resulting strengths. Estimates can reach the full 1,000 points; being absent from our list does not create a 970-point ceiling in the current scoring version.
The combination is 80% of the geometric mean of the clue strengths plus 20% of the weakest strength, multiplied by ten. A missing link therefore brings the score down sharply.
In bridge rounds, indefinite words such as “something” and broad evaluations such as “good” receive a specificity adjustment: their estimated strengths are multiplied by 0.6. These words fit many situations without identifying much of a connection. This adjustment never overrides a recognized answer and does not apply to close-meaning rounds. The result tells you when it was applied.
Where the language data comes from
The scoring vocabulary contains 100,000 dictionary-filtered words, represented by a compact version of Stanford’s GloVe 2024 WikiGiga vectors. Word forms and sense references use WordNet 3.0; its copyright notice and license accompany the data. The puzzles, meaning cues and connection explanations are authored separately.
Checking runs on our server. Your guess is not sent to an outside AI service, and no live language model invents a score.
An estimate still has limits.
Word-use patterns cannot capture every joke, cultural reference or unexpected connection. A sensible answer can still receive a disappointing estimate. The score is feedback within this game, not a judgment of your imagination.
Dates keep their original collection and scoring version. The October 2, 2026 edition and older shared practice links retain the earlier rules, including the 970-point ceiling on estimates. Current practice and the daily season beginning October 3 use the rules above.
Your strengths are not a leaderboard.
“80 / 100” describes a connection in this scoring system. It does not mean that you beat 80% of players. Your statistics come from games completed in this browser.
Put a little wordplay into your day.
Five rounds, one word at a time. Your next connection is waiting.
Take your seat