This dataset consists of approximately 5,000 round of Scrabble. For each round, I track the board state, the player rack, and the list of possible words according to a scrabble solver.
These were generated via self-play of gemini flash models as outlined in the ScrabbleBench article. Code for generating this data is available on Github.
