Klar.gg Chess

Chess engine assistance runs Stockfish (or another engine) locally on your device while you play a live game, displaying the engine's top recommended moves with evaluation scores in a browser overlay or separate window. At each move you see 2-4 suggested options ranked by evaluation, typically showing the top lines a few moves deep. You pick which move to make - the engine just informs the choice. The key configuration is engine depth: low depth (5-8) approximates 1400-1800 ELO consistency, medium depth (10-13) approximates 2000-2300 play, and high depth (15+) approaches GM/superhuman level. Running full depth on every move is detectable; using calibrated moderate depth is not. The overlay displays centipawn evaluation alongside move suggestions, which helps you calibrate which moves are critical (only large centipawn differences matter) versus essentially equivalent (several candidate moves within 10-20 centipawns of each other). In those equivalent positions you can safely play your preferred move rather than the engine's top choice, which both feels more natural and introduces the human variation that keeps your accuracy statistics undetectable.

Chess.com's detection system analyzes move accuracy compared against Stockfish recommendations across a game, looks for time usage patterns inconsistent with human thinking (humans slow down on complex positions; engine users often maintain consistent pace), checks correlation between your move quality and engine top choices across hundreds of moves, compares performance against your historical accuracy and tactical ratings, and flags unusual accuracy spikes in specific position types like endgames where engines excel far beyond typical human play. Lichess uses Irwin, an open-source model with similar methodology. Neither system detects the software itself - they detect the statistical fingerprint of engine-assisted play. Chess.com's Fair Play team reviews accounts flagged by the automated system before issuing a ban, meaning there's a human review step rather than purely automated enforcement. The review analyzes multiple recent games simultaneously looking for consistent patterns across different opponents, time controls, and positions - a single suspicious game rarely results in a ban, but a pattern of engine-consistent accuracy across 10-20 games triggers action.

A perfect Stockfish game at full depth produces near-zero centipawn loss per move - an immediate red flag for any detection system. Mistake injection deliberately introduces calibrated inaccuracies that match the error profile of a legitimate player at a specific rating. At 1500 ELO, a typical player makes several moves per game with 50-150 centipawn loss, occasionally misses tactics, and plays suboptimally in endgames. The injection system replicates this: it occasionally suggests the second-best move rather than the best, adds realistic thinking-time variance to simulate deliberation, and increases error rates in position types where your stated rating would realistically struggle. The result looks indistinguishable from a legitimately improving player. Position-type awareness in mistake injection is the most sophisticated component: legitimate 1600-rated players characteristically misplay rook endgames, struggle with passed pawn races, and miss multi-move tactical combinations involving more than two pieces. The injection system targets exactly these position types for its introduced errors while playing accurately in simpler tactical positions where a 1600 player would normally find the best move - creating a realistic profile rather than random noise.

The safest approach is matching engine depth to just above your current account's performance level. If you're at 1200 rated, use depth 7-9 to produce 1400-1600 quality play - a plausible improvement from consistent study. If you're at 1600, use depth 10-11 for 1800-2000 quality. Jumping from 1000 to 2500 ELO-level accuracy in a month is the clearest possible signal. Gradual improvement over weeks and months as you 'study' is what legitimate rating climbs look like. The depth control feature lets you set a per-game ceiling - use it to create a realistic performance trajectory rather than playing at your ceiling every single game.

The browser-overlay approach works on mobile browsers running Chess.com in desktop mode - Chrome and Firefox on Android with desktop mode enabled support the overlay. Native iOS and Android Chess.com apps run in isolated environments that prevent browser extension injection. The practical solution for iOS is running Chess.com in Safari with desktop mode requested, which allows the overlay with appropriate configuration. For regular mobile play the browser approach is more reliable than the native app approach on current iOS versions.

Manual review by Chess.com's Fair Play team is entirely analysis-based - they cannot see what's on your screen or running on your device. Their review examines the moves you played and compares them against engine recommendations statistically. With proper mistake injection and calibrated depth, your accuracy statistics across a game look identical to a legitimately strong player rather than engine-assisted play. The most common catch from manual review is players who use full-depth engine on every move without any injection - the near-zero centipawn loss across 40+ moves is unmistakable even to a human reviewer without any automated tool.