Per-Game Box-Score Projections
Per-batter and per-pitcher projections for every game in the projection window. Defaults to today's slate, top 50 by Total Bases (batters) or Strikeouts (pitchers). Methodology: Ridge regression per (role × stat) on a 5-feature substrate. Refreshed daily by the sports_lines cron.
Batter Projections
today's slate · sorted by total basesMethodology
ridge per (role × stat) · refreshed dailyInputs: 5-feature substrate (recent rate, season rate, opp baseline, park factor adjacency, role/lineup). Trained on prior seasons + in-season catchup.
Output grain: per (player × game) — i.e. an explicit row for each scheduled appearance in the window. Batters are projected across the full schedule (26,667 rows); pitchers are constrained to the near-term probable cohort (~78 rows / ~2-3 days), expanding in v2 to full rotation modeling.
What this is NOT: not a season-aggregate (use the team page for that), not an EV/odds layer (those live on the betting surfaces), not a starter-only filter (lineup composition matters — bench bats have rows too). The number is the model's best estimate of the per-game stat line.
v2 roadmap (banked): rotation modeling (expand pitcher cohort beyond probables), lineup × pitcher-handedness, park factors + weather (substrate exists), per-projection CI from Ridge variance + holdout MAE.