The bots of crypto,
ranked.

Methodology

Where the numbers come from. Every figure on this leaderboard is read from Polymarket's public Data API (data-api.polymarket.com) — the same backend that powers polymarket.com profile pages — plus market metadata from gamma-api.polymarket.com. For each tracked wallet we record: profit/loss and trading volume over the four windows Polymarket exposes (day, week, month, all-time) from /v1/leaderboard?user=<address>; open positions with unrealized P&L from /positions; realized P&L per closed market from /closed-positions; current portfolio value from /value; lifetime distinct-markets count from /traded; and full timestamped trade history from /activity. Every fill carries a Polygon transaction hash, so the activity behind each number is on-chain and independently monitorable. We always display the exact window next to a P&L figure. Polymarket does not publish a 90-day window; where we show one, we computed it ourselves from raw fills, label it as our estimate, and note it may diverge from Polymarket's own P&L accounting, whose internals (fees, mark-to-market, redemption timing) are not fully documented.

How wallets get classified as bots. Polymarket has no public flag for automated accounts, so classification is behavioral and probabilistic. We score each wallet on observable signals: round-the-clock activity (fills in 20 or more of 24 UTC hours across several days, with no sleep-like gap); sustained machine cadence (sub-2-second spacing between orders maintained over hundreds of orders, thousands of fills per day); unusual market breadth (thousands of distinct markets, spanning unrelated categories simultaneously); formulaic order sizing; volume far out of proportion to P&L, typical of market-making or arbitrage software; and public linkage to agent infrastructure, such as wallets funded from services in the Olas on-chain agent registry or matching open-source trading-agent frameworks. Each wallet receives a confidence grade — high, medium, or low — and we publish the specific signals behind every grade. When measuring cadence we count orders, not raw fills, because one order can fill against many counterparties in the same second.

What we do not claim. Classification is our judgment, stated with confidence levels — not a verified fact. A disciplined human using trading tools can look automated, and software executing one person's manual decisions blurs the line. We cannot prove any wallet is a person's or team's only wallet, and one operator may run many. Figures are P&L on Polymarket as computed by Polymarket, cached and refreshed periodically — not audited financial returns, and not evidence that performance will persist. Registry-based linkage is probable, not certain, and absence from a registry means nothing: most bots are private, unregistered scripts. The underlying data is public — anyone can re-run our checks against the same endpoints and hold us to them.

• Round-the-clock activity (strong)
• Sustained machine cadence (strong)
• Market breadth (strong)
• Volume grossly out of proportion to P&L (moderate)
• Reaction latency (moderate, designed metric)
• Formulaic order sizing (weak-moderate)
• Maker-side dominance (moderate)
• Agent-registry and funding linkage (strong when present, rare)
• Default profile metadata (weak, corroborating only)