Sudoku Clue-Count Benchmark Dataset

This dataset contains 32,000 standard 9 × 9 Sudoku instances, with 500 instances at every exact clue count from 17 to 80. Each instance includes its starting grid, unique completed grid, generation provenance and computational uniqueness evidence. The collection supports two distinct tasks: finding a valid completion and certifying uniqueness given one known completion. Clue count is the number of initially fixed cells and is treated as a controlled structural parameter rather than a direct measure of computational or human-solving difficulty. The benchmark supports investigating how fixed information interacts with clue placement, formulation and solver behaviour. The release provides 64 clue-count ZIP archives, a master instance index and a companion reproducibility archive containing generation and validation code, AMPL models and full solver logs. Each clue-count archive contains 500 individual CSVs with 81 cell rows recording row, column, given and solution. The 3,500 instances with 17–23 clues were constructed by transforming and augmenting existing source puzzles, while the remaining 28,500 instances were generated afresh by uniqueness-preserving clue deletion. Full source attribution and licence information are provided in the accompanying README.