Advanced stats — methodology
Stats on this site are computed empirically from our 65-game corpus across 2026, 2025, 2024, 2023. We mirror baseball.computer's formulas — same expressions, no MLB-published weights substituted in. The small-sample caveats below apply throughout.
The expectancy framework
- Run Expectancy (RE) per (outs, base state): the average number of runs scored from a given state to the end of the inning. Computed across our dataset.
- Win Expectancy (WE) per (inning, frame, margin, outs, base state): probability that the home team wins from a given state. Bayesian-smoothed with a broad prior (k=10) over a coarser key (margin in ±3 buckets, runners on/off).
- ΔRE per event = runs on play + RE(end) − RE(start). Sum per batter = wRAA. Sum per pitcher = run value above average (negated).
- ΔWE per event = WE(end) − WE(start) from the batting side's perspective. Sum per player = WPA.
- Leverage Index (LI) at a game state = avg |ΔWE| at that state ÷ league-average. By construction average LI = 1.0.
- pLI = a player's average leverage index faced. A pLI of 1.20 means the player was up in situations 20% above average importance.
Linear weights (2026)
Each play's run value, computed empirically. Two anchorings: wOBA-equivalent (mean-zero across all plays) and relative (in-play out anchored at 0).
| Play | n | Raw | wOBA-eq | Rel | flags |
|---|---|---|---|---|---|
| InsideTheParkHomeRun | 4 | 1.553 | +1.565 | +2.209 | imputed (n<30) |
| Triple | 30 | 1.225 | +1.237 | +1.881 | |
| GroundRuleDouble | 2 | 1.194 | +1.206 | +1.850 | imputed (n<30) |
| HomeRun | 1 | 1.078 | +1.090 | +1.734 | imputed (n<30) |
| Double | 107 | 1.038 | +1.050 | +1.694 | |
| ReachedOnError | 135 | 0.742 | +0.754 | +1.398 | |
| IntentionalWalk | 11 | 0.692 | +0.704 | +1.348 | imputed (n<30) |
| Single | 540 | 0.664 | +0.676 | +1.320 | |
| HitByPitch | 94 | 0.591 | +0.603 | +1.247 | |
| Walk | 345 | 0.541 | +0.554 | +1.197 | |
| Interference | 5 | 0.479 | +0.491 | +1.135 | imputed (n<30) |
| AdvancedOnError | 2 | 0.462 | +0.474 | +1.118 | imputed (n<30) |
| PassedBall | 32 | 0.273 | +0.285 | +0.929 | |
| WildPitch | 174 | 0.213 | +0.225 | +0.869 | |
| StolenBase | 306 | 0.178 | +0.191 | +0.834 | |
| Balk | 30 | 0.122 | +0.134 | +0.778 | |
| OtherAdvance | 65 | 0.020 | +0.032 | +0.676 | |
| FieldersChoice | 93 | -0.475 | -0.462 | +0.182 | |
| SacrificeHit | 8 | -0.604 | -0.592 | +0.052 | imputed (n<30) |
| StrikeOut | 457 | -0.633 | -0.621 | +0.023 | |
| InPlayOut | 931 | -0.656 | -0.644 | +0.000 | |
| SacrificeFly | 24 | -0.770 | -0.758 | -0.114 | imputed (n<30) |
| CaughtStealing | 37 | -0.881 | -0.869 | -0.225 | |
| PickedOff | 14 | -0.946 | -0.934 | -0.290 | imputed (n<30) |
Phase K product layer
The presentation layer builds on the expectancy framework above with Savant-style percentile cards and FanGraphs-breadth splits. Anything depending on pitch coordinates, exit velocity, or pitch types is not in this dataset — we have zone-aggregated batted ball location only. Every spatial visual is honest about that.
- Percentile cohorts: same-season qualifiers (≥50 PA for batters, ≥30 BF for pitchers) league-wide. Cells with fewer than 20 qualifiers in the cohort are grayed out in the UI.
- Splits: precomputed for pitcher hand, venue, base/ out state, leverage. RISP = runner on 2nd or 3rd. RISP 2-out narrows further to the 2-out subset. Leverage buckets: High (LI ≥ 1.5), Medium (0.7 ≤ LI < 1.5), Low (LI < 0.7).
- Count tendencies: 4×3 grid of (balls, strikes) states with per-pitch swing/contact/whiff rates. Cells with fewer than 10 pitches are dimmed; check the title text for the underlying count.
- Times through order (TTO): which PA against this pitcher in a given outing. 1 = first PA, 2 = second, 3+ = third or later. The dashed line on the TTO chart marks the pitcher's own 1st-TTO OPS baseline.
- Fatigue buckets: pitch count thrown before the PA started: 1-15, 16-30, 31-50, 51+. Resets per outing.
- Batter-vs-pitcher (BvP): career across all seasons in the dataset. With our ~5 games per season between any two teams, sample sizes are thin — surface as flavor, not as decision input below 20 PA.
- Park factors: basic only — runs per inning at the home park vs. away by the same team. With one venue per team in this league, the number partially captures team strength.
Sample-size caveats
- RE matrix per (outs, base state): ~200 events per cell. Stable.
- WE matrix: ~3,500 unique keys, most with n<5. The Bayesian broad-prior smoothing dominates for thin cells, so WE values stay well-behaved.
- Linear weights per (season, play): cells with n<30 are imputed from the cross-season global average. Flagged in the table above.
- Park factors: with one team per park, basic PF partially captures team strength, not just park geometry. Treat as preliminary until we ingest more seasons.