Bot Detection

How we estimate whether a wallet is human or automated — and why it matters.

Why It Matters#

A significant portion of Polymarket volume comes from automated trading bots. When studying trader behavior or comparing performance, it's useful to know whether you're looking at a human making decisions or a bot executing a programmatic strategy. Bots are flagged with a badge on all leaderboard tabs and trader profiles.

Heuristic Factors#

Bot probability is estimated from four weighted behavioral signals:

Trade Frequency (40% weight)#

Normalized fill rate per active trading day. Bots tend to trade at rates that would be impractical for a human — dozens or hundreds of trades per day, sustained over long periods. Wallets with extremely high sustained frequency are auto-classified as bots.

Maker Ratio (30% weight)#

The percentage of fills from passive limit orders versus aggressive market orders. A high maker ratio suggests automated market-making — posting limit orders on both sides and profiting from the spread.

Activity Entropy (25% weight)#

Shannon entropy of the hourly trade distribution (UTC). Maximum entropy means uniform 24-hour activity with no sleep pattern — a strong indicator of automation. Human traders show irregular gaps for sleep, meals, and daily life. Wallets with very high entropy and enough trading history are auto-classified as bots.

Size Regularity (5% weight)#

The inverse coefficient of variation of trade sizes. High regularity means consistent sizing typical of algorithmic execution. Humans tend to vary position sizes based on conviction, while bots use fixed or formulaic amounts.

Hard Rules#

Some wallets are auto-classified as bots (score = 100%) regardless of the composite score when they trigger a definitive rule:

• High Frequency — sustained high fill rate over many active days
• Uniform Activity — very high hourly entropy with sufficient trading history

Wallets that don't trigger a hard rule need a minimum number of trades for composite scoring. Below that threshold, the score defaults to 0%.

Score Interpretation#

• 0–20% — likely human
• 20–50% — some automated characteristics
• 50–80% — probably automated
• 80–100% — almost certainly a bot

Known Limitations#

• Sophisticated bots that deliberately randomize their behavior may score low
• Professional human traders using hotkeys and alerts may score higher than expected
• The score is a probability, not a binary classification — there is no hard threshold
• New wallets with few trades have less data, so scores are less reliable