Free during beta —Free during beta — to track favorite players + get game alertsto track favorites + alerts
Free tool · MLB
Matchup Predictor
Pick any two MLB teams. The model's win probability for each — from team Elo ratings, home-field included. Works year-round.
@
BOS 45%55% TB
The model makes Tampa Bay Rays a 55% favorite at home — Slight lean.
BOS · Elo 1573TB · Elo 1587
Win probability from the model's team Elo ratings, home-field included. This is the model's read on the matchup — not a betting line, and not a claim to beat the market. How accurate is the model? See the Calibration Grade →
How it works
Each team carries an Elo rating updated after every game by the same model that powers our public track record.
The win probability is the Elo expectation for the matchup, with MLB's measured home-field edge applied to the home side.
It's the model's honest read — how often it expects each side to win — not a betting line and not a market-beating claim.
Elo compresses every past result into one rating per team, weighting recent games and margins. The win probability comes from the standard /400 curve applied to the rating gap plus home-field advantage — a disciplined summary of results so far, updated after every game.
It is deliberately blunt: no injuries, no rest, no travel, no lineup news. That bluntness is what makes it consistent — and what makes it a starting point rather than a verdict.
Worked example
Home team rated 1580, visitor 1520, home advantage worth 55 points: effective gap 115. The /400 curve gives 1 ÷ (1 + 10^(−115/400)) ≈ 66%. Flip the venue and the same two teams land near 58% — home field alone is worth roughly eight points of probability in this range.
When NOT to use it
Single games with big news — a starting QB out or a rest night isn't in the rating yet. Elo learns from results, which means it learns late.
Early season: ratings carry over from last year (regressed), so September numbers describe last season more than this one.
Don't difference this number against a sportsbook line and call the gap an edge. We grade our model in public — including a closing-line value that is negative — precisely so nobody has to take that claim on faith.