Methods
Every number, explained.
A high-level overview of what the daily pipeline considers. We describe the inputs, not the weights.
Feature groups
Batter signal
- Season and recent-window batting performance, plus deviation from the season baseline.
- Batted-ball quality indicators such as barrel rate and expected outcomes.
- Lineup slot and expected plate appearances for the day.
Opposing pitcher
- Season and recent-start performance, pitcher tier, and handedness matchup.
- How hittable the starter has been, and how often he allows home runs.
- Prior head-to-head history when the sample is meaningful.
- Strikeout profile, used to build the 3+ through 10+ K ladder for each starter.
Game and totals
- Starting-pitcher quality on both sides, plus recent bullpen workload and effectiveness.
- A run-distribution model for the total, rather than a single point estimate.
- Recent team form, including extended win and over/under streaks.
- Where available, published market prices are used as a sanity check on our own number.
Context
- Park effects, including park-specific home-run tendencies.
- Weather.
- Umpire tendencies.
Research overlay
- A separate numerology alignment board is published alongside the models.
- It is a research-and-entertainment overlay and does not drive the statistical projections.
Models & calibration
- Probabilities are calibrated against graded outcomes and re-tuned on a rolling window, so published numbers track how the model has actually been performing.
- Game markets are additionally sanity-checked against published prices where those are available.
- Every daily slate is locked before first pitch, graded once games are final, and shown on the History page — hits and misses.
- Past results do not guarantee future outcomes. Every pick can lose.