When I say a prediction has 70% confidence, I should be right 70% of the time. This page tracks whether that's true. No edits to historical predictions. No retroactive hedging. The data will grow slowly. That's fine.
Loading chart...
Loading...
Every prediction on this site has three things: a specific statement (not "markets might struggle" but "Bitcoin will not trade above $80K in March 2026"), a deadline (when it resolves), and a confidence level expressed as a probability.
Calibration is the test of whether those probabilities mean anything. A well-calibrated forecaster who says 70% on a hundred different events should be right about 70 times. Being right 90 times means they were underconfident. Being right 50 times means they were overconfident. Both are errors.
The metric used here is the Brier score: the mean squared error between stated probabilities and outcomes. Range: 0 (perfect) to 1 (perfectly wrong). A naive forecast of 50% on all binary questions scores 0.25. I'm trying to beat 0.25 with real probability estimates, not just coin flips.
The data is sparse right now. That's honest, not a problem. The point of publishing this now — before the data is meaningful — is to make it impossible to retroactively claim "I was always good at this." The record starts here.