Viral quant strategy claim

Prediction markets let traders buy contracts that pay $1 if an event happens and $0 if it does not, so a 60-cent price is usually read as a 60% chance. Polymarket says prices reflect the market’s collective belief, and Congress’s research service describes event-contract prices the same way in liquid markets. (docs.polymarket.com) (congress.gov) The viral claim takes that basic idea and adds a trading rule: estimate a “true” probability, compare it with the market price, and buy only when the gap is large enough. Open-source bots on GitHub now advertise that exact workflow, scanning hundreds of binary markets, estimating fair odds with Anthropic’s Claude and other models, and sizing positions with the Kelly criterion. (github.com 1) (github.com 2) The Kelly criterion is a bankroll formula that tells a trader how much to bet when they think they have an edge. One of the GitHub projects says it scans “hundreds of binary markets” every 10 minutes, while another paper-trading bot sets a 10% minimum edge, a 10% cap per trade, and a five-position limit. (github.com 1) (github.com 2) That helps explain why the thread spread: it turns a familiar Wall Street idea into a retail prediction-market recipe with code people can inspect. GitHub search results now show multiple repositories built around Polymarket, Claude, and “mispricing” detection, including bots for weather, politics, and general event markets. (github.com 1) (github.com 2) (github.com 3) The hard part is the phrase “true probability.” In finance markets tied to stock prices or weather, outside data can anchor that estimate; one recent project says it uses weather ensemble forecasts and trades only when model probability differs from market price by more than 8%. (github.com) (github.com) In politics, geopolitics, or breaking-news markets, the estimate often comes from model judgment over public information rather than a clean formula. A Claude-based Polymarket bot on GitHub says it searches recent news, has Claude 3.5 Haiku generate fair probabilities, and then paper-trades signals against live market prices. (github.com) Even if a model finds a price gap, execution and settlement can erase it. Polymarket’s own documentation says traders need to read each market’s resolution rules, because the title is not enough; markets resolve against specific sources, can be disputed, and unresolved edge cases can end in a 50-50 payout. (docs.polymarket.com) The market itself can also move before an order fills. Polymarket’s data documentation shows best bid and offer prices, full order-book depth, and streaming market data, which means a scanner may spot a theoretical edge that disappears once spreads, slippage, and thin liquidity are accounted for. (docs.polymarket.us) Academic and policy research gives a mixed backdrop. A January 2026 paper on Kalshi, the federally licensed United States prediction market, found that contract prices become more accurate as markets approach closing but still show a favorite-longshot bias, with low-price contracts winning too rarely to break even. (cepr.org) (karlwhelan.com) That leaves the viral performance numbers in a narrower category than the thread suggests: they may describe one backtest, one paper-trading run, or one favorable three-month window, but not a verified market-wide law. The code for AI-driven scanners is easy to find; independently audited records for the headline returns are not. (github.com) (github.com)