These numbers describe how closely play matched a chess engine, compared with human players at the same rating. They are not a finding that anyone cheated. Strong players, prepared openings and simple positions all produce engine-like play. Please don't publish or act on accusations based on this page.
Check a game
Check a player
One game rarely shows anything. Looking at many games by the same player is far more informative.
How it works
Every position in a game is searched by Stockfish 17.1 in your own browser, with fixed settings (300,000 nodes, five candidate moves, one thread, cleared memory before every search). The same settings were used to analyse the human games in the baseline, so the numbers can be compared directly. Nothing you analyse is sent anywhere.
Which moves count
Forced moves, the first 8 moves of each side (opening theory) and moves made when the game was already decided (an evaluation of 5 pawns or more) are left out. The rest are "eligible".
Critical moves
A move is critical when the best move is at least 10 percentage points of winning chances better than the second-best. Selective engine use, a few moves per game, tends to show up here first.
Accuracy and losses
Evaluations are converted to winning chances (the Lichess formula) before losses are worked out, so one missed mate in a won position cannot distort a game. Accuracy uses the Lichess per-move formula.
Engine matches compared with expectation
For every eligible move, the baseline gives how often human players of the same rating, in the same time control, matched the engine in positions of similar difficulty. The z-score compares the player's actual matches with that expectation: around 0 is typical, and larger values mean more matches than expected. It treats moves as independent, which they are not quite, so its real false-positive rate comes from the validation work, not from a normal table.
Clock use
"Time against difficulty" is the rank correlation between thinking time and how much the best move stood out. People usually think longer when one move matters, so values near zero or below are less common. "Time on critical moves" compares median thinking time on critical moves with other moves.
Baseline
Human games from the Lichess open database (CC0), grouped by 200-point rating band and time control. Percentiles place a game among baseline games from the same group.
Limits
A single game rarely supports any conclusion. Strong players, home preparation and simple positions all produce engine-like play. The platforms see browser, device and behaviour signals that a PGN file does not contain. Bullet clock data is dominated by premoves.