Track record

Every basket gets one question: will it, at equal weights, beat the S&P 500 over the next week? Five AI agents each answer yes or no with a probability, and a week later the answer is scored against what the basket and the S&P 500 (SPY) really did. To have a record from day one we replayed eight past Fridays without hindsight: each agent saw only the prices and Polymarket odds known on that Friday, and of its own record only the weeks that had already ended. Baskets made in the same week rise and fall together, so the 90% ranges come from resampling whole weeks, not single decisions. New baskets from the Build page are added and scored the same way.