Polymarket bot decoded; backtest tools shared
A deep dive decoded a top Polymarket sports‑betting quant bot that reported $7.4M PnL and described tactics like buy‑only engines, Dutching, volume farming and triangular hedging. Separately, an open‑source NautilusTrader Python/Rust backtesting framework with Brier‑advantage metrics for prediction markets was published. (x.com) (x.com)
Prediction markets work like stock markets for yes-or-no questions: a share priced at $0.63 implies a 63% chance, and traders can buy or sell before the event ends. A new pair of open-source posts pulled back the curtain on how some of the fastest Polymarket sports traders try to exploit those prices. (docs.polymarket.com) (github.com) One thread by 0xMovez described a sports-betting bot that it said generated $7.4 million in profit and loss on Polymarket, then broke the system into tactics including buy-only engines, Dutching, volume farming and triangular hedging. The post was published on X and framed as a reverse-engineering of one of the venue’s top sports-market quant traders. (x.com) A second post by Atenov_D pointed readers to a public GitHub repository for prediction-market backtesting built on NautilusTrader, an open-source trading engine written in Rust with Python bindings. The repository says it supports Polymarket and Kalshi through custom adapters and was updated again on April 12, 2026. (x.com) (github.com 1) (github.com 2) The mechanics matter because prediction markets are not house-set odds boards; they are order books where traders can briefly push linked outcomes out of line. Polymarket says prices reflect the market’s collective belief, and that structure leaves room for bots that can react faster than humans. (docs.polymarket.com) (finance.yahoo.com) Academic researchers have already documented that pattern at scale. An August 2025 study summarized by Yahoo Finance said researchers analyzed 86 million Polymarket bets from April 2024 to April 2025 and found nearly $40 million in arbitrage profits, with the top three wallets placing more than 10,200 bets for $4.2 million in profit. (finance.yahoo.com) (arxiv.org) In plain terms, Dutching means spreading stake across several outcomes so the payout is balanced if any one of them wins. Triangular hedging means combining positions across related markets so a trader can lock in a small edge when the combined prices do not add up cleanly. (arxiv.org) (docs.polymarket.com) The new backtesting code tries to answer a different question: whether those ideas would have worked on historical data after fees, slippage and queue position. The GitHub readme says the framework includes equity curves, fill charts, drawdowns, monthly returns and a running “cumulative Brier advantage,” a forecasting metric that compares probability calls with actual outcomes. (github.com) NautilusTrader’s own documentation says the engine is designed so the same strategy logic can run in research, simulation and live trading with the same event-driven model. That matters for prediction markets because many paper profits disappear when a live order reaches the book later than a backtest assumed. (nautilustrader.io) (github.com) Polymarket’s sports section now spans thousands of active markets and more than $816 million in trading volume, according to the platform’s sports page. In a market that large, public code and public teardown threads give smaller traders a closer look at how professional-style bots measure edge before they risk real money. (polymarket.com)
