Track Record separates cautious forecasts from higher-variance research. Accuracy appears after recorded markets resolve.
A cautious view that favors stronger evidence and smaller, more defensible probability gaps.
Compare predicted probabilities with observed outcomes.
| Category | Resolved | Accuracy | Avg Edge vs Market |
|---|---|---|---|
| No resolved forecasts yet. | |||
| Date | Market | Category | AI Pred | Mkt Price | Edge % | Side | Outcome |
|---|---|---|---|---|---|---|---|
| No resolved forecasts yet. | |||||||
Unlike simple win/loss tracking, Continuous Ranked Probability Score (CRPS) measures how close our probability estimates are to the actual outcome. A forecast of 90% on an event that happens scores better than a forecast of 55% on the same event — even though both are technically "correct." This incentivizes precision, not just being on the right side.
Every night, our system reviews all resolved forecasts and computes per-category calibration multipliers. If we're systematically overconfident in crypto forecasts (e.g., predicting 80% when the true rate is 75%), the multiplier corrects for this bias going forward. This process runs automatically and the results are reflected in the calibration chart above.
When all 6 models agree (6/6), our confidence is highest. When models disagree significantly (3/6 or 4/6), we flag lower confidence and adjust our calibrated output downward. Disagreement is itself a valuable signal — it often indicates genuinely uncertain markets where the outcome is harder to predict.