How it works

Every number on a results page comes from a process built for one job: spotting engine assistance in chess. There's no language model anywhere in it. Just a chess engine, measurable features of every move, and a machine learning model we trained ourselves.

Now live

Custom AI-powered cheat detection

No LLMs. Our machine-learning model was built from scratch for one task: detecting cheating in chess. The latest version is trained on thousands of real grandmaster games, so honest strong players are 16× less likely to be mistaken for an engine.

Model
Custom ML v3, built from scratch
LLMs
None
Inputs
Per-move engine and positional features
Rating-aware
Judges moves against human play at your rating
Target
Engine assistance, not move quality
Status
Live

The process

Seven steps take a game from a link or a file to a verdict for each player.

  1. 01

    Providing a game

    Paste the game's PGN text, upload a .pgn file, or paste a Chess.com or Lichess link and we'll fetch the game for you.

  2. 02

    Reading the game

    We read the game into its parts: the players, their ratings, the date, the result and every move played.

  3. 03

    Engine evaluation

    Stockfish, one of the strongest chess engines, checks every position on our servers, so we know the best options at each point in the game.

  4. 04

    Per-move details

    For each move we measure how close it came to the engine's best options: whether it was the top choice, in the top few, and how much it gave away.

  5. 05

    Human move prediction

    We estimate how likely a player at that rating would be to find each move. A move that's routine at 2200 can be a rare find at 1200.

  6. 06

    Custom ML scoring

    The core of the detector

    Our own model weighs all of these signals together, along with both players' ratings, and gives each player a suspicion score. It's trained to recognize engine assistance, not to grade how well someone played.

  7. 07

    Engine match timeline

    Every results page shows the whole game on one chart. Taller bars mean a closer match to the engine, and colors run from common to rare moves for that rating. Blue marks scored moves when ratings are unknown; grey marks moves that aren't scored.

    • Common
    • Plausible
    • Uncommon
    • Rare
    • No rating data
    • Not scored

Ready to try it?

Paste a PGN, upload a file, or load a Chess.com or Lichess game. No account needed.