Proven Edge
Every number here is measured over graded calls — never simulated, never back-filled. The platform reports; it never trades. Judge it on the record.
HOW WE GRADE
Each call is a directional prediction that resolves the way the model said or it doesn’t — measured against the realized move, not a claim. A vertical is proven only when its Wilson-90% lower bound (the honest worst case) clears a 50% coin flip; the p-value is the chance the record arose by luck if the model had no edge. Verticals with fewer than 20 graded calls are flagged building-sample. Value-betting sources (which win under 50% by design) are scored by profit, not hit-rate, and excluded here.
Directional accuracy over graded calls — did each call resolve the way the model predicted? Bounded [0,100%], unit-clean, comparable across verticals. Hit-rate floor is the Wilson 90% lower bound (honest worst-case); Brier scores how well the calibrated win-prob matches reality (lower is better). A research lens on model quality, not a P&L claim — the platform never trades. Verticals with under 20 decided calls are flagged building-sample.
Measured over the trailing 365 days · updated 8/13/2026, 8:31:56 AM