Model Calibration

Scoring Methodology

Calibration score = (1 − (P(yes) − outcome)²) × 100 — Brier-based probability accuracy, where outcome is 1 for YES / 0 for NO. A model that assigns low probability to an event that doesn't happen scores high.

Cascade fidelity = (confirmed + 0.5 × partial) ÷ predicted effects × 100 — only applicable when the modeled premise materialized (market resolved YES); shown as n/a otherwise.

Prospective models only — scenarios run before market resolution. Retrospective scenarios are excluded from the average score.