Viral quant‑model leak thread
A viral social thread claims a Citadel intern revealed a four‑factor quant model—cross‑market divergence, disposition coefficient, capital velocity and pair network correlation—and shows a rebuilt bot allegedly winning 70% on prediction‑market data, though the posts note skepticism about the claim. The thread has drawn large engagement and debate over reproducibility and the ethics of sharing proprietary strategies. ( )
A viral X thread says a Citadel intern leaked a four-part trading model, but the public posts offer no independent proof of the leak or the bot’s results. (x.com) Quant trading means using code and statistics to hunt for tiny pricing mistakes across many markets at once. Citadel says its Global Quantitative Strategies unit trades more than 15,000 securities algorithmically across more than 30 countries, and its interns work on real projects and statistical models. (citadel.com, citadel.com) The thread describes four signals in plain terms: gaps between related markets, signs traders sell winners too early and hold losers too long, how quickly money rotates through positions, and how tightly linked contracts move together. Those ideas map to familiar quant concepts like cross-market divergence, behavioral bias, turnover, and correlation networks, but the posts do not publish a verifiable research note or source document. (x.com) A second post says someone rebuilt the model into a prediction-market bot and showed about a 70% win rate. The claim spread into a wider debate over whether the number came from live trading, paper trading, or backtesting on old data. (x.com) That distinction matters because prediction markets are easy to simulate and harder to verify in real time. Polymarket presents itself as the world’s largest prediction market, and guides for bot verification rank on-chain or exchange-logged trades above screenshots and backtests because historical tests can be overfit. (polymarket.com, agentbets.ai) Public code repositories show how quickly these claims can turn into copycat tools. One GitHub project advertises a Polymarket trading toolkit with paper-trading strategies, and another prediction-market bot repo promotes a 67.3% demo win rate built around machine-learning features and strict entry and exit rules. (github.com, github.com) What is still missing is the part that would settle the argument: a primary document, a named source at Citadel, or a wallet or exchange log tying the alleged rebuild to real trades. Until that appears, the thread stands as a fast-moving social-media claim about a secret model, not a verified account of one. (agentbets.ai, citadel.com)
