The interesting part is in the splits.
A research notebook for the questions behind the box score: counts, innings, pitch sequences, fatigue, leverage, and changing styles. Some views are explorable today; others are clearly marked as future data work.
Research sources are linked. Illustrative research examples below are externally published aggregates—not reconstructed player-level results from our current database.
What happens when the lineup turns over?
Published MLB figures for starting pitchers, combined 2016–2025. This example is not yet calculated from our own pitch-by-pitch records.
Source and interpretation
MLB's Times Through the Order Penalty glossary reports the 2016–2025 averages for starts. The pattern is descriptive; pitcher quality, batter quality and which starters remain in games affect the comparison. Individual pitchers need larger samples and selection corrections.
Is it fatigue—or familiarity?
Pitchers face a selectively tougher sample when they remain deep in a start. Which part comes from pitch count, which from seeing the same hitter again, and which from managers pulling struggling pitchers?
OPS allowed, first to third trip, 2016–2025.
A 2026 study explicitly modeled the competing explanations and found that adjustment and selection assumptions matter. This is the sort of deeper analysis we'd like to make explorable.
Pitch/event rows with game ID, hitter, pitcher, pitch count, times-through-order, count, game state and outcome.
One pitch can reshape the plate appearance.
Source: MLB's catcher-framing glossary. This is an outcome conditioned on *reaching* a count and should not be mistaken for the causal effect of a single called strike.
Every 0–0 to 3–2 state, on one map.
Tap any cell to see league outcomes, then filter by hitter, pitcher, handedness, pitch type, year, and ballpark. A second lens shows pitch-level decisions: swing, take, chase, foul, contact, whiff.
Cells are navigation concepts, not invented historical OPS values. Actual values are withheld until event-level ingestion.
Questions worth answering
Filter by what data is required. "Available" means the concept can be derived from existing source tables, not necessarily that its exact view is already implemented.
A graphic and its receipt travel together.
The model for every future finding: an immediately readable number or graphic, a one-sentence interpretation, the exact population and window, and a permalink into the underlying analysis.