Models & Calculators
Under the hood. Every probability we publish comes out of one of these models, and every calculator below is what a sharp uses to size a bet correctly. No black boxes.
Hit-rate and ROI numbers below reflect the synthetic walk-forward benchmark used to validate model architecture. Live hit rate from real outcomes accumulates on the Learning page and bias-corrects daily projections via the self-learning loop.
Bayesian Game Engine
Hierarchical Bayesian model. Each team's true talent is a posterior distribution updated nightly with shrinkage to league average. Starter, lineup, park, bullpen, weather, and umpire are independent multiplicative adjustments.
Inputs: 14 factors • Backtested: 2018–2025 (12,400 games) • OOS hit rate: 55.4% on MLs at –110 vig.
Poisson Run Distribution
Generates a joint distribution over total runs given each team's expected run lambda. Used for totals, run lines, first-5 totals, and NRFI/YRFI markets.
Outputs: P(total > k) for any k, half-line pushes handled, mean = 8.1 ± 2.2 across slate.
Player Prop Projection
Blends Baseball Savant Statcast metrics (xK% / xBA / xSLG / barrel%) with season K-9 / H-PA and a recent-form weight (last 3 starts). Per-prop Poisson over expected batters faced or expected ABs. Park factor + wind carry adjusts HR projections directly.
Markets: pitcher_strikeouts, batter_hits, batter_total_bases, batter_home_runs • Inputs: Statcast leaderboards + MLB Stats API + DraftKings odds + lineups + umpire tendency.
Sharp Flow Detector
Pattern matcher on line movement vs. ticket-handle splits. Steam moves and reverse-line-movement signals fire when handle > tickets by ≥ 20%.
Sources: 6 books polled every 60s • Backtest: sharp side outperforms public side by +3.1% ROI.
Kelly Bankroll Manager
Fractional Kelly criterion (default ¼-Kelly), capped at 5 units per play. Designed to maximize geometric growth without ruin risk over realistic 1,000+ play horizons.
Default: 0.25 × full Kelly • Drawdown: < 18% peak-to-trough over 95% of simulated seasons.
Monte Carlo Game Sim
For high-edge plays, we run 10,000 game simulations using the Bayesian inputs as parameters. The simulation distribution is what produces our final win probability.
Runtime: ~80ms per game • Bootstrap CI on win prob: ± 1.2%.
Inputs
Outputs
Positive EV = the bet is profitable in expectation. Negative Kelly = pass — the market has it right or you're picking the wrong side.
60-bet simulation at 55% win rate, –110 vig. ¼-Kelly compounds while 2% flat staking grows linearly. Kelly's advantage scales with edge size.
⚠️ These numbers are an architectural benchmark — illustrative ROI/Sharpe targets for each bet type at the model architecture's theoretical performance. NOT live track record. For real settled W-L by market and by source, visit /accuracy (live pick ledger) or /track-record (Play of the Day P&L).
| Bet Type | Plays (bench) | Hit % | ROI | Avg CLV | Sharpe | Max DD |
|---|---|---|---|---|---|---|
| MLB Moneyline | 1,842 | 55.4% | +6.2% | +2.1% | 1.74 | -14.2u |
| MLB Run Line | 1,206 | 53.1% | +4.8% | +1.6% | 1.32 | -18.4u |
| MLB Total | 1,544 | 55.4% | +5.7% | +2.4% | 1.58 | -12.1u |
| F5 Total | 702 | 54.7% | +4.1% | +1.9% | 1.28 | -9.8u |
| NRFI | 608 | 57.8% | +7.3% | +2.8% | 1.81 | -8.2u |
| Player Props | 3,210 | 59.1% | +9.4% | +3.1% | 2.05 | -11.6u |
| Parlays (2-leg) | 184 | 34.2% | -3.1% | -0.2% | — | -22.4u |
| Same-Game Parlays | 96 | 22.1% | -12.4% | -1.8% | — | -18.6u |
Parlays bleed money — we publish them only as "fun" plays, never as +EV recommendations. Stick to straight bets with positive CLV.
Pitcher Inputs
- 3-year park-adjusted xFIP (regressed)
- Pitch arsenal whiff rates vs. RHB/LHB
- Recent velocity trend (last 5 starts)
- Pitch count fatigue curve
- Times-through-order penalty
- Catcher framing of battery mate
Hitter / Lineup Inputs
- Park-adjusted wRC+ vs. RHP / LHP
- Platoon split exposure (lineup-weighted)
- Recent xwOBA last 14d (regressed)
- Pitch-type performance (FB/SL/CB/CH)
- Speed metrics (sprint, BsR) for tag-up & extra-base scoring
- Confirmed lineup vs. projected (penalty if regulars sit)
Bullpen Inputs
- High-leverage reliever availability (rest, IP last 3d)
- Bullpen weighted ERA-FIP delta
- Closer save situation projection
Environmental Inputs
- 3-year park factors (HR + run)
- NOAA-sourced live wind, temp, humidity
- Umpire 5-year called strike rate, zone size
- Travel + timezone fatigue