Evaluating RSO-F

RSO-F is being publicly evaluated on ForecastBench, which tests predictions against events that have yet to happen. Results are available on its preliminary dataset leaderboard and are updated as questions resolve.

The preliminary dataset leaderboard covers resolved questions based on real-world time series. Prediction-market questions are evaluated separately; these results are not an overall ranking across both categories.

We study where forecasts perform well, where they fail, and how reliably their probabilities reflect what happens.

Each new application needs its own evaluation.