Methodology · MLB
How the MLB model works
Playoff, division, pennant, and World Series odds for all 30 clubs.
- Version
- v1 (Pythagorean + log5, regression 0.25, + deadline overlay)
- Updated
- Daily
- Data source
- MLB Stats API (statsapi.mlb.com) — free, no key; documented roster overlay
- Calibration
- Game predictions locked pregame and graded nightly on the Ledger
The question
Each club's odds to reach the playoffs, win its division, win the pennant, and win the World Series — plus a projected final win total.
The data
The free MLB Stats API: current standings, runs scored and allowed, and the full remaining schedule. Pulled every day in season.
How strength is measured
Talent starts from the Pythagorean expectation — a club's expected win rate from runs scored and allowed (exponent 1.83), which predicts future record better than raw win-loss. It's then regressed 25% toward .500 to account for luck.
Recency & regression
The rating uses season-to-date run differential, so it updates every day as games are played; the 25% pull toward average keeps a hot or cold month from being taken at face value.
Injuries & roster
Talent is inferred from team run differential, which already reflects who's been playing — but it can't see a trade the day it happens. So a documented overlay handles deadline moves: a dated, reasoned win adjustment (e.g. +4 for a contender that lands an ace), applied over the remaining schedule with a written reason on every entry. No player-level modeling beyond that.
Home advantage
A fixed home-field edge (+0.035 to the home win probability) is added on top of the log5 matchup.
The simulation
For each pair of clubs, log5 turns the two talent ratings into a game win probability. A 10,000-run Monte Carlo plays out every remaining game to get final standings, then simulates the bracket: Wild Card (best-of-3) → Division Series (best-of-5) → LCS and World Series (best-of-7).
Postseason format
12 teams, six per league: the three division winners plus three wild cards. The top two seeds in each league get a first-round bye.
Backtesting
Every day the model locks a pregame win probability for each game and grades it against the result — a public, running calibration and Brier score you can watch build on the Ledger.
Known weaknesses
No starting-pitcher, bullpen, or lineup modeling; talent is run-differential only; and injuries enter only through their effect on results, not directly.
Every published number is timestamped to the Ledger the moment it's made and graded against results — nothing here is backdated. Percentages are rounded for readability.