
thumb|A simple KenKen puzzle, with answers filled in as large numbers. KenKen and KenDoku are trademarked names for a style of arithmetic and logic puzzle invented in 2004 by Japanese math teacher Tetsuya Miyamoto, who intended the puzzles to be an instruction-free method of training the brain. The name derives from the Japanese word for . The names Calcudoku and Mathdoku are sometimes used by those who do not have the rights to use the KenKen or KenDoku trademarks.
thumb|A simple KenKen puzzle, with answers filled in as large numbers. KenKen and KenDoku are trademarked names for a style of arithmetic and logic puzzle invented in 2004 by Japanese math teacher Tetsuya Miyamoto, who intended the puzzles to be an instruction-free method of training the brain. The name derives from the Japanese word for . The names Calcudoku and Mathdoku are sometimes used by those who do not have the rights to use the KenKen or KenDoku trademarks.
==General rules== As in Sudoku, the goal of each puzzle is to fill a grid with digits –– 1 through 4 for a 4×4 grid, 1 through 5 for a 5×5, 1 through 6 for a 6×6, etc. –– so that no digit appears more than once in any row or any column (a Latin square). Grids range in size from 3×3 to 9×9. Additionally, KenKen grids are divided into heavily outlined groups of cells –– often called “cages” –– and the numbers in the cells of each cage must produce a certain “target” number when combined using a specified mathematical operation (one of addition, subtraction, multiplication or division). For example, a linear three-cell cage specifying addition and a target number of 6 in a 4×4 puzzle must be satisfied with the digits 1, 2, and 3. Digits may be repeated within a cage, as long as they are not in the same row or column. No operation is relevant for a single-cell cage: placing the "target" in the cell is the only possibility (thus being a "free space"). The target number and operation appear in the upper left-hand corner of the cage.
Discovered by embedding cosine similarity (sentence-transformers MiniLM, 384-dim).