Arcade · five playable algorithms

Arcade

Five games, each running its actual solver — the same module the game page loads — so what you see move is the search itself, not an animation of one.

Playable systems
0
Provable guarantees
0
Median 2048 score
—
Third-party deps
0

The five systems

  1. 2048

    A table-driven engine — all 65,536 row transforms precomputed — under a depth-adaptive expectimax search over chance nodes.

    • Measured, not guaranteed
    • Expectimax
    • Transposition table
    Play 2048

  2. Connect Four

    Iterative-deepening negamax with alpha-beta pruning and a Zobrist transposition table. It reports forced wins as proofs, with the move count.

    • Proves forced wins
    • Negamax
    • Alpha–beta
    Play Connect Four

  3. Snake

    A Hamiltonian cycle with guarded shortcuts. This is the one AI here with a real proof: it cannot die, and it fills all 144 cells.

    • Provably always wins
    • Hamiltonian cycle
    • Invariant
    Play Snake

  4. Minesweeper

    Constraint propagation, then subset subtraction, then exact enumeration of every consistent mine arrangement to get the true odds.

    • Deductions are proofs
    • CSP
    • Exact probability
    Play Minesweeper

  5. Sudoku

    Knuth's Algorithm X with dancing links, plus a generator that re-solves after every removal to guarantee a unique answer.

    • Solves any valid grid
    • Dancing links
    • Exact cover
    Play Sudoku

On guarantees

“Can the AI always win?” has a different answer for each of these. Randomness and hidden information are what break a guarantee — not the quality of the search.

Game Guarantee Why
Snake always wins Deterministic and fully observable. The cycle invariant makes a fatal move unreachable.
Sudoku always solves A finite exact-cover search. If a solution exists, Algorithm X finds it.
Connect Four perfect at the horizon Solved in theory, but proving it from an empty board is expensive. Inside its search horizon the result is exact.
Minesweeper no Hidden information. Some boards reduce to a genuine coin flip; the solver reports the odds instead of bluffing.
2048 no Tiles spawn at random, so for any policy some spawn sequence beats it. Only a measured rate is honest.

Measured 2048 strength

Games sampled
—
Reached 2048
—
Reached 1024
—
Median score
—
Mean moves
—

Produced by scripts/benchmark-2048.mjs. Run npm run test:2048 to reproduce these numbers locally — the benchmark drives the same engine module this page loads, over seeded games, so a run is reproducible.

The studio

Back to the room it all began in.

Every page here is on the computer in that room.

Enter the studio