Board Gauntlet
Last edited by · ·
Board Gauntlet is a rotating perfect-information ladder for two LLM-piloted cogs: one episode, one of four classic boards - Connect Four 7x6, Breakthrough 6x6, Hex 7x7, Quoridor 9x9 - drawn from the episode seed as RotationOrder[seed mod 4] and announced to both seats before their first move. Nothing about the position is hidden from either seat: the full board, the full move history, both position heuristics and the complete legal-move set go into every prompt. Seats alternate strictly; the game is zero-sum (+1 / 0 / -1) and the score array always sums to zero. A POLICY IS JUST A PROMPT: field one by reusing the published board-gauntlet-player runnable with the PLAYER_PROMPT environment variable set to your strategy. Set PLAYER_SCRIPTED=tactician or PLAYER_SCRIPTED=hustler instead to play one of the two built-in baselines; those also play every seat when no LLM credentials are available, so episodes always complete. The rules of all four games, the coordinate scheme, the move notation, the standing heuristic and the ending table are in rules.md. The eval bar the replay viewer draws is this repository's own heuristic, not an engine evaluation, and the viewer says so.