COUNCIL RULING: MLB Player Props Model Anchor Methodology
Date: 2026-04-03
Ruled by: Boss (Ernest)
DECISIONS:
- Career data source: MLB Stats API (free). Pull career totals per player.
- Execution threshold: Tier C (track only) for now. Data collection phase - no executable edges. Will reassess when sample accumulates.
- H2H threshold: 30 PA minimum, max 20% weight. Below 30 PA use pitcher archetype matching.
- Kalshi microstructure: Not building into edge calc now. Boss will evaluate liquidity/spreads manually.
- Career prior: MLB Stats API career stats as primary. Steamer/ZiPS for rookies later if needed.
KEY METHODOLOGY ADOPTED:
- Per-PA rates (not per-game) as foundation
- Bayesian shrinkage with research-backed k values (HR=170PA, K=60PA, BABIP=800+PA)
- Log-Odds opponent adjustment
- Gaussian copula for HRR correlation
- Two-component NegBin mixture for Pitcher Outs
- 3-year handedness-specific park factors
- Pitcher archetypes for opponent matching (below 30 PA H2H)
- Dual-anchor tracking with DM test + mSPRT
- Confidence tier system (currently Tier C)
DEFERRED:
- Ridge regression betas (need training data)
- Monte Carlo RBI/Runs simulation
- Spring training data integration
- Steamer/ZiPS projection integration
Source: ~/edgeclaw/data/councils/2026-04-03/mlb-prop-anchor-model/ruling.txt