Loading...
6 Specific Clues We Check
Instead of relying on one mystery accuracy number, we check six everyday habits — hover or tap any ? badge to see what each clue looks for.
Catching "Smart" Part-Time Cheating
Tested on unseen games where assisted players kept their overall accuracy inside normal human ranges so standard site checks missed them:
| Hidden Help Style | Basic Accuracy Check | Stealth Clue Score | Full v5.2 Score | Caught |
|---|---|---|---|---|
| Loading test results... | ||||
Multi-Game Player Profile Results
Tested on 55 complete player profiles (4–6 games each) never seen during training. Across all 32 genuine human players (from club level to 2890+ Grandmasters), 0 out of 32 were falsely flagged for priority review:
| Method Compared | Separation | Precision | Caught | Error |
|---|---|---|---|---|
| Loading profile test results... | ||||
How Chess.Observer Works — in 4 Simple Steps
1. Human Moves vs. Engine Moves
Playing Stockfish's #1 move isn't suspicious when it's an obvious recapture or check. We compare every turn against a model of what real humans at that rating naturally play, spotting moves that only a computer would love.
2. Spotting Part-Time Help
Players who only check an engine twice a game or peek at a win/loss bar keep their average accuracy completely normal. We look specifically for sudden clutch-turn spikes, zero blunders in tense moments, and #2-move disguises.
3. Reading Move Rhythm Fairly
Humans play obvious moves fast and pause on hard puzzles; copying from a second screen creates a flat 2–6 second delay on every turn. Fast bullet games (<2 min) are automatically excluded so quick mouse skills are never punished.
4. Helping Humans Decide
No algorithm should ever accuse someone on autopilot. Check at least 4–6 games, use the Player Pool slider to account for honest hot streaks, and treat every flag as a starting point for fair human review.