Baseball · Models
Moneyline tree
The active win-probability head: what it predicts, how it was trained, how it scores, and which explanations are available.
What it predicts
Home-team win probability
A calibrated classifier estimates which team wins. The active system release determines how this probability contributes to recommendations and sizing.
Recorded component metadata
Some component metadata is absent from the published evidence. The verified system identifiers are on the overview; missing fields are not inferred.
- Model version
- retrosheet-game-date-bounded-fielding-2026-v16.0.0
Historical Tree benchmark
Evaluation
Forecast quality against the market
Historical out-of-sample evidence for the historical Tree cohort (retrosheet-game-date-bounded-fielding-2026-v16.0.0). Brier and log loss are forecast-error scores; lower is better for both.
No model-specific benchmark artifact has been published yet.
Moneyline feature importance
Tree training feature importance
Ranked by how much split gain the moneyline model gave each stat during training. “Validation” checks the stat on games the model did not train on: helped is encouraging, hurt means the stat needs more scrutiny. Attention alone does not prove a stat predicts winners.
- Stats considered
- 72
- Training passes
- 73
- FanGraphs stats
- 32.3% split gain
- Model run
- mr_baseball_v16_0_0_20260829
- #1Game dataMatchup edge: expected winning percentage from runs scored and allowed19.62% split gain · used in 19 model rules
The home team's expected winning percentage from runs scored and allowed minus the away team's. A positive value means the home team has more of this statistic. Whether that is favorable depends on the statistic and the model; higher ERA or xFIP is generally worse.
- #2Game dataMatchup edge: season run differential13.02% split gain · used in 13 model rules
The home team's season run differential minus the away team's. A positive value means the home team has more of this statistic. Whether that is favorable depends on the statistic and the model; higher ERA or xFIP is generally worse.
- #3Game dataMatchup edge: start-of-day margin above or below the wild-card cutoff9.50% split gain · used in 15 model rules
The home team's start-of-day margin above or below the wild-card cutoff minus the away team's. A positive value means the home team has more of this statistic. Whether that is favorable depends on the statistic and the model; higher ERA or xFIP is generally worse.
- #4FanGraphsMatchup edge: the lineup's prior-year wRC+9.47% split gain · used in 27 model rules
The home team's the lineup's prior-year wRC+ minus the away team's. A positive value means the home team has more of this statistic. Whether that is favorable depends on the statistic and the model; higher ERA or xFIP is generally worse.
- #5Game dataMatchup edge: runs allowed per game this season5.49% split gain · used in 16 model rules
The home team's runs allowed per game this season minus the away team's. A positive value means the home team has more of this statistic. Whether that is favorable depends on the statistic and the model; higher ERA or xFIP is generally worse.
Showing the top 5 of 72 features from .
All 72 training features
| Feature | Training gain share |
|---|---|
| Matchup edge: expected winning percentage from runs scored and allowed | 19.62% |
| Matchup edge: season run differential | 13.02% |
| Matchup edge: start-of-day margin above or below the wild-card cutoff | 9.50% |
| Matchup edge: the lineup's prior-year wRC+ | 9.47% |
| Matchup edge: runs allowed per game this season | 5.49% |
| Matchup edge: the starter's prior-year xFIP (lower is better) | 5.38% |
| Matchup edge: manager career winning percentage | 4.86% |
| Matchup edge: the starter's prior-year strikeout-minus-walk rate | 4.82% |
| Matchup edge: the starter's season-to-date strikeout-minus-walk rate | 4.56% |
| Matchup edge: the lineup's prior-year defensive runs | 3.69% |
| Matchup edge: wins above replacement pace per plate appearance | 1.92% |
| Matchup edge: start-of-day playoff-race leverage | 1.90% |
| Matchup edge: the starter's season-to-date xFIP (lower is better) | 1.80% |
| Matchup edge: wins above replacement this season | 1.71% |
| Team-strength edge (venue-neutral) | 1.53% |
| Matchup edge: the lineup's prior-year wOBA | 1.48% |
| Matchup edge: batters the starter has faced this season (evidence weight) | 1.38% |
| Matchup edge: proximity to the start-of-day playoff qualification cutoff | 1.37% |
| Matchup edge: the starter's ERA over the last 30 days | 1.12% |
| Matchup edge: difference between pitcher pitch-quality results and the lineup’s results against that mix | 1.07% |
| Combined level: the starter's prior-year xFIP (lower is better) | 0.61% |
| Matchup edge: runs allowed per game over the last 10 games | 0.54% |
| Matchup edge: the starter's strikeouts per nine this season | 0.52% |
| Combined level: offensive production against the opposing starter hand after broad team quality | 0.47% |
| Matchup edge: the starter's offspeed-pitch usage share | 0.32% |
| Matchup edge: record-vs-run-differential luck factor | 0.26% |
| Matchup edge: the lineup's season-to-date wOBA | 0.21% |
| Matchup edge: the bullpen's WHIP over the last 7 days | 0.19% |
| Combined level: high-leverage closer outs recorded over the last 7 days | 0.18% |
| Matchup edge: the lineup's season-to-date defensive error prevention | 0.16% |
| Matchup edge: lineup results against breaking pitches matched to opposing breaking-ball usage | 0.16% |
| Combined level: the lineup's prior-year wOBA | 0.15% |
| Combined level: runs allowed per game over the last 10 games | 0.15% |
| Matchup edge: the bullpen's ERA over the last 7 days | 0.14% |
| Combined level: bulk-relief outs recorded over the last 7 days | 0.13% |
| Matchup edge: the starter's ERA this season | 0.12% |
| Matchup edge: high-leverage closer outs recorded over the last 7 days | 0.00% |
| Combined level: share of recent bullpen work handled by closer-type relievers | 0.00% |
| Matchup edge: share of recent bullpen work handled by closer-type relievers | 0.00% |
| Combined level: the bullpen's ERA over the last 7 days | 0.00% |
| Combined level: bullpen fatigue risk | 0.00% |
| Matchup edge: bullpen fatigue risk | 0.00% |
| Matchup edge: bullpen innings worked over the last 3 days | 0.00% |
| Combined level: middle-relief outs recorded over the last 7 days | 0.00% |
| Matchup edge: middle-relief outs recorded over the last 7 days | 0.00% |
| Combined level: setup-relief outs recorded over the last 7 days | 0.00% |
| Combined level: the lineup's prior-year defensive runs | 0.00% |
| Matchup edge: the lineup's season-to-date wRC+ | 0.00% |
| Combined level: the lineup's prior-year wRC+ | 0.00% |
| Matchup edge: manager career games (the evidence behind the record) | 0.00% |
| Matchup edge: manager team winning percentage this season | 0.00% |
| Matchup edge: offensive production against the opposing starter hand after broad team quality | 0.00% |
| Matchup edge: same-handed lineup alignment with the opposing starter’s pitch mix | 0.00% |
| Matchup edge: opposing starter pitch usage concentrated in the lineup’s weaker pitch buckets | 0.00% |
| Matchup edge: lineup results against offspeed pitches matched to opposing offspeed usage | 0.00% |
| Combined level: runs allowed per game this season | 0.00% |
| Matchup edge: runs scored per game over the last 10 games | 0.00% |
| Run-scoring park factor | 0.00% |
| Matchup edge: the starter's breaking-ball usage share | 0.00% |
| Combined level: the starter's ERA over the last 30 days | 0.00% |
| Matchup edge: the starter's fastball usage share | 0.00% |
| Combined level: the starter's prior-year strikeout-minus-walk rate | 0.00% |
| Combined level: the starter's offspeed-pitch usage share | 0.00% |
| Combined level: the starter's pitch-mix diversity rather than reliance on one pitch family | 0.00% |
| Matchup edge: the starter's pitch-mix diversity rather than reliance on one pitch family | 0.00% |
| Combined level: the starter's WHIP this season | 0.00% |
| Matchup edge: the starter's WHIP this season | 0.00% |
| Matchup edge: recent travel burden from distance, direction, time-zone shift, altitude gain, and road chain | 0.00% |
| Combined level: whether prior venue history was available to measure travel burden | 0.00% |
| Matchup edge: winning percentage over the last 10 games | 0.00% |
| Matchup edge: winning percentage over the last 30 games | 0.00% |
| Matchup edge: expected wins above replacement this season | 0.00% |
Per-game Tree contribution vectors are not yet published. Open a game notebook to inspect the retained inputs and forecast.