Polymarket quant bot backtested at scale

Prediction markets turn a yes-or-no question into a tradable price, and a viral post this month said one Polymarket bot used that math to test 72.1 million trades. (xrticles.com) On Polymarket, each market is a binary contract with Yes and No outcomes, and the platform displays the midpoint between best bid and best ask as the implied probability when spreads are narrow. The public order book shows bids, asks, spreads, and historical prices through Polymarket’s trading application programming interface. (docs.polymarket.com 1) (docs.polymarket.com 2) That setup lets traders look for mispricing, meaning a contract price that differs from their estimate of the true odds. Expected value is the core test: if a trader’s probability estimate implies a positive average return over many bets, the trade is attractive in theory. (oddschecker.com) (docs.polymarket.com) The post described a bot that combined mispricing formulas, expected-value filters, and the Kelly Criterion, a bankroll rule that sizes bets as a fraction of capital rather than going all-in. The article tied those rules to a backtest across 72.1 million trades and $18.26 billion in volume. (xrticles.com) (theory.stanford.edu) Kelly sizing is a long-run growth formula first published by John L. Kelly Jr. in 1956, and it increases stake size when the estimated edge is larger. In plain terms, it is a way to bet more when the math looks better and less when the edge is thin. (en.wikipedia.org) (app.olympus-bets.com) Automated trading matters on these venues because prices and liquidity update continuously, and Polymarket’s own documentation steers developers toward real-time streams instead of repeated polling for production use. Polymarket US says its institutional market-data stream is designed for lower-latency access to best bid and offer data. (docs.polymarket.com) (docs.polymarket.us) The thread landed into a market already crowded with bot builders, open-source code, and no-code products aimed at scanning dozens or hundreds of contracts for edges. GitHub repositories and commercial tools now advertise Kelly sizing, edge detection, and automated execution for Polymarket-style markets. (github.com 1) (github.com 2) (predictengine.ai) The caution is that backtests are not live trading, especially in markets with spreads, slippage, fees, and changing liquidity. Polymarket’s documentation shows spreads and minimum order sizes vary by market, which means a paper edge can shrink once real orders hit the book. (docs.polymarket.com) So the post’s real claim was narrower than “a bot beats the market”: that prediction-market prices can be turned into a large-scale data set, and that simple rules can be tested against it. Whether those rules survive live execution is the question traders still have to answer one order at a time. (xrticles.com) (docs.polymarket.com)