High forecast confidence — every pick graded
Games the model gave 70%+ to win — itemized with the actual result. A forecast-confidence tier, not a market-aware recommendation. ~1–2 a night.
What this is
Every day our model rates each MLB game with a win probability — an Elo + gradient-boosted (XGBoost) ensemble trained on the 2021–2025 seasons (13,048 games). The "pick" is the side it favors. We grade that pick against the actual final score from MLB's official stats feed. Nothing is cherry-picked — the model's full daily output, every game, every day.
High-conviction = games where the model is ≥10 points off a coin flip.
The honest part
This is straight-up winner accuracy, not profit. Favorites win ~55% by base rate, so the number alone isn't an edge. Whether picks make money at real prices is a different question — that's the Recommendation record tab, which grades only published qualified selections at their exact quoted odds, flat one unit. Some days hit 80%, some 27%. The variance below is real, and we show all of it.