mine-sm

A deterministic, no_std 8×8 Minesweeper state machine for reinforcement learning. This page is a browser harness: play the board, inspect the agent-facing observation and action mask, auto-play a guaranteed winning or losing line, and run the conformance checks that validate this JS mirror against the Rust engine. Source & docs.

Click to act; right-click to flag. Flag mode flips left-click.
State: playing Moves: 0 Strikes: 0 Flags: 0 Revealed: 0/64

Auto-play (ground truth, never guesses)

The winning line repeatedly opens the lowest hidden safe cell; the losing line opens a numbered cell then a bomb.

Observation (agent input)

−1 hidden · −0.5 flagged · −2 detonated · 0.0–0.8 = adjacent bombs / 10

Action mask (1 = legal reveal)

Conformance checks

    Not run yet.