MLB Game Desk — Model Anchor Calibration Prompt

Context: Our MLB Player Props council (2026-04-03) established the universal framework: per-possession/per-PA rates as foundation, Bayesian shrinkage toward career prior, Log-Odds opponent adjustment, Gaussian copula for correlated combo props, dual-anchor system (sportsbook + model) tracked via Brier scores, confidence tiers. This prompt asks for the SPORT-SPECIFIC parameters to plug into that framework.

MLB Game Desk — Model Anchor Calibration

We are building model anchors for MLB game-level betting markets on Kalshi: moneylines, run spreads, totals (over/under), first-5-innings, and team totals.

Questions requiring sport-specific recommendations:

1. What team-level metrics best predict run scoring? (Pythagorean expectation, BaseRuns, team wRC+/FIP, run differential?)
2. How should starting pitcher quality drive the total and spread? SP xFIP/SIERA vs ERA? How much does SP identity explain vs team offense?
3. How many team games until team offensive/defensive metrics stabilize? What k values for team-level Bayesian shrinkage?
4. How should the model generate a total (O/U)? Team projected runs = f(team offense, opposing SP, park, weather)?
5. How should the model generate a spread? Run differential from team ratings + SP adjustment?
6. First-5-innings modeling: how does this differ from full-game? SP dominates F5, bullpen irrelevant.
7. Team totals: how to split the full-game total into home/away team components?
8. Bullpen quality and availability — how should it adjust full-game totals but not F5?
9. How should home-field advantage be quantified? Park-specific or universal?
10. Weather adjustments for totals (wind, temperature, humidity, altitude)?
11. Umpire impact on totals and F5?
12. What distributions? Totals=Poisson? Spread=Normal?
13. EWMA alpha for team offensive/defensive metrics?
14. Early-season stabilization: when do team-level metrics become trustworthy vs preseason projections?
15. Give specific numbers, formulas, and implementation-ready recommendations.