Where the numbers come from

Every morning the pipeline pulls the full PrizePicks MLB board, then builds its own view of each prop from real data: season stat rates for every player on the slate, posted batting orders and probable pitchers, park factors, sportsbook game totals, and player-prop lines from major books.

The projection

Hitters are projected as rate × opportunity — a hitter's per-at-bat rates scaled by the plate appearances his lineup spot earns (a leadoff bat gets ~4.6 trips, a 9-hole bat ~3.8). Pitchers use per-inning rates times expected innings. The projection is then nudged for the matchup: ballpark, the opposing starter's WHIP/K-rate, the platoon hand matchup, and the betting market's run total for the game.

The blend

The model's probability is blended with two reality checks: the player's last-10-game hit rate on that exact line, and the sportsbook consensus with the bookmaker's margin stripped out. When the sharps disagree with us by a lot, the pick gets downgraded — the market is the strongest single signal in sports.

Grading & calibration

Every pick is archived and graded against the box score the next morning — wins, losses and no-plays all stay on the Record tab, nothing is quietly deleted. As graded picks accumulate, an isotonic calibration layer learns how often a “75%” really hits and corrects future probabilities automatically. Each row is stamped with the model version so eras never get mixed.

Tiers & lineups

CORE needs the model, recent form and the market to all agree. FLEX and WATCH are progressively looser; LEAN is informational. Lineups headline the slip-level win estimate — the chance every leg hits, which is much lower than any single leg — and an EV estimate that accounts for goblin/demon payout differences.

Estimates, not betting advice. Probabilities are model outputs with real uncertainty; payout multipliers are approximations — always verify the live PrizePicks line and payout before entering anything.