Live (placed)

+33.99u

509 bets · 18.32% ROI

Backtest (counterfactual)

+112.02u

425 bets · +41.07% ROI

Staked (counterfactual)

272.73u

walk-forward Kelly

Why the gap? The live ledger reflects what the model recommended at the time it ran — config, thresholds, and model artifacts as they existed each day. The backtest uses today's training + today's policy applied walk-forward (no future-data leakage by fold construction). Recent pipeline fixes (injury contamination, side-flip floors, tier-mask sweeps) are reflected in the backtest but apply only forward-going in the live ledger.
Important — do not bet to the backtest number. Backtest ROI is structurally higher than live ROI in this system because (a) the backtest evaluates a single candidate model while production serves a stacked ensemble whose distribution differs, (b) model and policy selection over the same history introduces survivorship bias even with walk-forward folds, and (c) the backtest assumes the recommended price is fillable, which is only sometimes true at game-time. Treat the backtest as a sanity check on signal direction, not as a forecast of forward ROI. The live ledger is the system of record (master.md Rule 9); this strip is informational only.

Backtest coverage in this window

● MLB Baseball 234 bets (through 2026-04-30) · MLS 11 bets (through 2026-04-12) · NBA Basketball 44 bets (through 2026-04-30) · NHL Hockey 127 bets (through 2026-04-29) · US Open Cup 9 bets (through 2026-04-29)
● Not in backtest: NFL Football, WNBA Basketball, College Football, College Basketball (M), UFC / MMA, Serie B, MLB_EXTRAS, NBA_PROPS, MLB_PROPS, NHL_PROPS, WNBA_PROPS, NFL_PROPS — backtest folds for these sports are blocked by the locked OOS holdout (WNBA: 2026 season is all inside the 30-day holdout; UFC: step×validation math caps folds ≈7 weeks behind today). Their live bets are counted on the left strip but contribute 0 to the backtest comparison until folds catch up. Live tracking is the only forward signal for these sports right now.