Rubik's Cube — a solver that shows its decisions

A real 3D cube: drag to orbit, press a face to turn it, scramble it as deep as you like. Then ask for the layer-by-layer plan — cross, corners, middle edges, then orientation and permutation, step by step with a coach line explaining each. The solver is deterministic and every move is verified by simulation before it is played. When several setups reach the next phase in the same number of moves, the tie goes to the engine — and the rung chip on each step says who picked it. The seed is optional: leave it empty and every game rolls a fresh random number, shown so the scramble stays reproducible.

What decides what

  • The method is a rule, full stop. Layer-by-layer: cross, first-layer corners, middle edges, edge orientation, corner orientation, corner permutation, edge permutation, final alignment. None of that is ever “asked” — it is the fixed skeleton of the solve, and every candidate insert is verified against the 54-facelet model before it is played.
  • The genuine ties are the set-ups. Sune or anti-sune, Ua or Ub, U-twist before or after: when several alignments reach the next phase in the same number of moves, the engine arbitrates. The step's rung chip reads rule (no tie), roster (tie, nobody asked), or engine (tie, the runtime chose).
  • A refusal cannot derail the solve. If the engine declines or is off, the solver takes the first equal option — roster order — and the rung says so. The plan still verifies.
  • Determinism is test-pinned. Same seed, same scramble, same plan — and the pinned PLL algebras (pure edge 3-cycles, double swaps, corner transpositions) are asserted by simulation in the test suite.

Honest edges

  • The solver is beginner layer-by-layer, not a speed solver: expect ~140 moves, not 20. The point is a plan you can read, not a record.
  • Turns are instant with a face flash, not tweened layer animations — the state you see is always the exact facelet state, with no animation drift between what is shown and what is solved.
  • A hand move mid-plan drops the rest of the plan on purpose: a plan computed for a state you then changed is a guess wearing a lab coat. Press solve again for a fresh one.
  • The scoreboard is session-only: close the tab and it is gone. Nothing leaves this page.