Methodology
The data, calculations, and publishing process behind Mallitalytics.
The daily product
A full MLB slate involves the schedule, probable starters, recent player performance, and the prior day's results. Mallitalytics brings these pieces into a daily briefing, with links to games and player profiles.
The homepage collects starter projections, recent hitter and pitcher signals, and completed-game highlights. Cards and season leaderboards provide player detail. Stories develop individual baseball questions through analysis.
Data sources
- MLB game feeds: schedules, probable starters, game status, box scores, and play-by-play records.
- Statcast: pitch-level tracking, including velocity, movement, location, exit velocity, launch angle, and expected outcomes where available.
- Mallitalytics calculations: summaries, rolling comparisons, MalliScore, and model-generated starter projections prepared from those records.
From data to publication
- Prepare the data. Python processes game and pitch records into reusable datasets for player analysis and daily reports.
- Build the outputs. Those datasets supply the briefing, projections, cards, and season profiles. Preparing summaries ahead of publication avoids repeating full-season calculations for every website visit.
- Check and review. Automated checks inspect expected outputs and publication dates and flag failures. MalliHub, the internal workspace, coordinates schedule preparation, card generation, output checks, content review, approval, and publishing.
- Publish. Routine data updates run on a schedule. Editorial content is reviewed before publication. Publication times and data cutoffs identify which update readers are seeing.
Game feeds and pitch tracking can become available at different times. A completed box score may appear before all tracking-based statistics have refreshed.
Reading player statistics
Check the date and sample shown with a statistic. A game line describes one game; a season line covers the season to its stated cutoff. Recent skill signals compare the last 21 days with the player's earlier season. They are rolling-window comparisons, not changes since yesterday.
Hits, home runs, runs, RBI, and OPS describe recorded results. Exit velocity, barrels, and expected statistics describe other aspects of performance. xwOBA estimates offensive value across plate appearances. xwOBAcon uses only tracked balls in play, which is the contact-quality view used on hitter cards and homepage contact leaderboards.
Whiff rate uses swings, swinging-strike rate uses pitches, and chase rate uses pitches outside the strike zone. Missing tracking can reduce the available sample. Short windows and small samples can move sharply after a few games.
MalliScore
MalliScore is a 0-100 grade for a completed start, combining dominance, run prevention, and workload. It summarizes that outing. The MalliScore article explains the construction and interpretation of the scale, including the 60+ elite-start category.
Starter projections
The projection board estimates outcomes for listed probable starters using information available before the game. Daily updates refresh inputs such as pitcher performance and opponent context. The trained model stays fixed between separately evaluated model updates.
- Projected MalliScore: the expected outing grade.
- Ace chance: the estimated probability of a MalliScore of 59 or higher.
- Bust chance: the estimated probability of a MalliScore below 30.
- Expected line: projected innings, strikeouts, and the other displayed pitching statistics.
The projection model's 59+ ace threshold differs from the completed-score article's 60+ elite category. The starter projection article covers the model and its outputs in more detail.
Updates and corrections
“Data through” identifies the source-data cutoff. A publication timestamp identifies when an output was generated or published. A newer publication time does not necessarily include another day of games.
Official scoring changes and later tracking updates can revise published statistics. Report a specific issue through Contact, including the page and number involved. See Corrections for the editorial correction process.
