How the Suspicious Move Score Works

The method, step by step

1. Reconstruct the position before the move. We start from the board as it stood immediately before your opponent's most recent turn, by removing the tiles they just placed.
2. Analyze legal, strong moves with a solver. A game solver evaluates the best plays available from that position, giving us a reference for what optimal or near-optimal play looks like.
3. Compare against human gameplay patterns. The observed move is measured against how human players actually play similar positions — because “strong” and “unusual” are not the same thing.
4. Account for unknown rack tiles. You rarely know your opponent's full rack. Where necessary the model samples plausible racks so the result reflects a range of possibilities rather than one assumed hand.
5. Calculate the Suspicious Move Score. These signals are combined into a single 0–100 score with a plain-language classification.

Why compare against both solvers and humans

A move that a solver loves is only suspicious if humans rarely find it. Many strong plays are also popular plays — they are the obvious move, and most players would choose them. Those score low. The score climbs when a move is simultaneously close to optimal and rarely selected by human players, especially when that happens repeatedly.

This is why a single high score is weak evidence on its own. Luck and skill both produce excellent individual moves. A sustained pattern of highly solver-like moves is what becomes statistically difficult to explain by unaided human play.

What the score is — and isn't

The Suspicious Move Score is an analytical signal, not a verdict. It does not observe intent, does not access anyone's device, and cannot prove that a person used a solver. Use it to decide where to look more closely, ideally across multiple games rather than one move.