The knowability of the future: Evidence from one million collective forecasts

How much of the future can humans know? Decisions across science, policy, and economics rely on probabilistic forecasts, yet the empirical limits of prospective knowledge remain unclear. Using approximately one million resolved prediction-market contracts across four platforms, twelve domains, and time horizons ranging from hours to years, we evaluate the calibration of collective forecasts in real-world settings. We find that market-implied probabilities are highly calibrated: events assigned probability p occur at approximately p percent frequencies. Calibration emerges early, when information is scarce and uncertainty remains high, and persists across platforms and domains despite systematic biases. These results suggest that large-scale aggregation of human judgment yields probabilities reliably reflecting outcome frequencies, indicating that the future may be statistically knowable earlier and more broadly than often assumed.