Track Record — Real Settled Picks

Realized cumulative P&L across every settled pick, 1u flat. Blended across all sources (including experimental prop feeds), so it can run negative even while the sharp sources profit — see the per-source breakdown below.

Peak-to-trough drawdown of the cumulative units curve over the settled history.

Closing line value is the #1 leading indicator of long-term profit. Even when you lose a play, if you beat the closing line you're playing a winning game. Per-pick CLV capture is being wired in (write-once open + closing-line snapshots); average CLV shows "—" until enough real closing lines accumulate.

Every day, the model is allowed to learn from prior days only. No look-ahead. This is the gold-standard backtest for time-series strategies. Backtests that train on the whole dataset and evaluate the same data ("in-sample") overstate performance by 2-5x.

Same backtest engine runs nightly in CI on the real season. Any month with ROI <0 or hit % <52 triggers an auto-recalibration of the model's prior assumptions. Bayesian priors are weighted ~50 game shrinkage.