1/ In 1906, Francis Galton asked 787 fairgoers to guess the weight of an ox. Their average was within 1% of truth. For 120 years, science, meteorology, and finance followed this rule: when in doubt, average everyone. Until now. We are solving the failure mode of the crowd. Follow live: zenodivergent.dev/live-read 🧵 👇

" For over a century — dating back to Francis Galton’s 1906 observation of the crowd at the county fair — the bedrock rule of forecasting across meteorology, economics, intelligence, and markets has been the same: when in doubt, average everyone together. The wisdom of the crowd is notoriously hard to beat, because individual biases cancel out. Until they don't. When the crowd agrees for the wrong reasons — shared blind spots, common training data, herd behavior, or correlated models — the average isn't wise; it's confidently wrong. This moment is the first time anyone is testing whether a neural foundation model can learn how consensus breaks across completely different domains. In our previous run, we saw the first concrete proof that this is possible: - Wikipedia event outcomes: 28% lower error than the crowd average, and 14% better than a tuned gradient-boosted tree baseline. - Prediction markets: 39% error reduction over the plain average. - Physical weather: 6% error reduction over our previous production reference... "