Deterministic, no_std 3x3 color-cube state machine for reinforcement learning: 18-move API, cubie/facelet encodings, action masks, seeded uniform scrambles, plus Thistlethwaite and Kociemba two-phase solvers.