Inside the BotEdge Scoring Engine: How We Detect Polymarket Bots
With hundreds of millions of dollars flowing through Polymarket daily, separating retail traders from automated algorithms has become crucial for market research. At BotEdge, we don't rely on self-reported data or simple volume metrics.
Instead, we use a custom behavioral engine that scores wallets on a 0-100 scale of "bot-likelihood". Here is exactly how our scoring algorithm works.
The 6+1 Heuristic Engine
When an active wallet is detected, BotEdge analyzes its last 100-200 trades against 6 live behavioral heuristics. Each heuristic returns a score from 0-100 and has a different weighting in the final calculation. For qualifying wallets, we can optionally add a 7th latency review when matching price context is available.
>_1. Trade Frequency (Weight: 1.5)
Humans sleep. Humans take breaks. Bots operate 24/7.
- What we measure: Average trades per hour and median inter-trade interval.
- Bot signal: Wallets executing >10 trades per hour with median intervals under 5 minutes score extremely high (90-100).
- Human signal: Less than 2 trades per hour.
>_2. Timing Regularity (Weight: 1.5)
Bots are bound by cron jobs or loops, resulting in hyper-consistent execution timing.
- What we measure: The Coefficient of Variation (CV) of inter-trade intervals.
- Bot signal: A low CV (< 0.3) indicates machine-like regular spacing between trades. We also check 24-hour histograms to detect the absence of human sleep cycles.
>_3. Position Sizing Consistency (Weight: 1.2)
Human traders size their bets based on intuition and emotion. Algorithms use math.
- What we measure: The variety and variance of trade sizes (in USDC).
- Bot signal: A wallet with hundreds of trades but only 3-4 unique, highly specific trade sizes (e.g., exactly $42.50 every time) or sizes adhering strictly to Kelly criterion scaling.
>_4. Hold Time Consistency (Weight: 1.0)
The most common and profitable algorithmic strategy involves short-term mean reversion or scalping.
- What we measure: The duration between a BUY and subsequent SELL on the same market.
- Bot signal: Tight clustering of hold times. If 90% of trades are held for exactly 60 to 180 seconds, the wallet receives a near-perfect bot score for this heuristic.
>_5. Market Selectivity (Weight: 0.8)
Humans are distractible and trade across multiple categories (e.g., sports, politics, crypto). Focused bots stay in their lane.
- What we measure: The diversity of traded market categories.
- Bot signal: Exclusive focus on a single asset (like BTC 5-minute markets) despite opportunities elsewhere.
>_6. Win Rate Stability (Weight: 1.0)
A human's win rate fluctuates wildly based on luck and tilt.
- What we measure: Rolling 50-trade win rate windows.
- Bot signal: A win rate that stays confined within a highly specific, tight band (e.g., 68-74%) across all rolling windows, indicating a statistically robust edge rather than a lucky streak.
>_7. Response Latency (Weight: 1.3)
This is our most resource-intensive and most definitive test, so it is not part of every scoring pass. We only run it on wallets that already look suspicious and only when we have matching price context available.
- What we measure: The time delta between a significant move on the Central Limit Order Book (CLOB) and the wallet's trade execution.
- Bot signal: Consistent trade execution within milliseconds to a few seconds of a price shock.
The Final Classification
The scores from these heuristics are combined into a weighted average:
- 0-30 (Human): Normal retail behavior.
- 31-55 (Likely Bot): Semi-automated, UI-assisted, or hybrid strategies.
- 56-100 (Definite Bot): High-confidence systematic traders. These are the wallets most likely to be promoted into the featured BotEdge Leaderboard.
When a newly discovered wallet crosses the 56-point threshold, it becomes eligible for automated Telegram alerts so researchers can inspect it in real time.
