Balanced Attack Draw

The backtest replays six seasons of MLS data (2020-2025) through the exact same filter logic, odds join, and settlement rules used in live trading — so reported P&L reflects what would have been executed in production, not an idealized sum of per-signal returns. Odds come from actual closing-line snapshots across multiple US books, with book priority ordering applied to select the sharpest available price; returns are net of vig by construction. All rolling features are computed with one-game lag shifts to prevent look-ahead bias, and the backtest engine applies live portfolio controls (daily risk cap, drawdown protection, dedup) so individual-strategy results aren't inflated by unrealistic assumptions. Walk-forward validation splits the sample into a training window (2020-2022) and an out-of-sample test window (2023-2025) — the strategy must perform comparably on unseen data to survive into active deployment, and filter thresholds are frozen before the test period to prevent parameter tuning on the holdout.

Self-reported. Not independently verified.