Live predictions

Tonight's games

Every prediction carries the timestamp it was written at. Nothing here is scored yet, which is the point — a forecast is only evidence if it existed before the result did.

loading…

How to read this

The bar is always the home team's win probability, so it is comparable across every game; the tick marks even money. Below 50% means the model is picking the road team.
The PICKS column names the team the model favours and its chance of winning. That is the call the record is scored on. Baseball's ceiling is low: even perfect knowledge of team quality caps single-game accuracy near 57–60%.
pre-game means the prediction was timestamped before first pitch and counts toward the record. provisional means it was not, and does not count.
The two names are the announced starters. The +3.7 pp beside them is how far those pitchers move the home win probability, in percentage points.
Predictions may move until first pitch as lineups and starters firm up, then they are locked. The number shown after a game begins is the one that stood when it began.
53–54% is home-field advantage alone, 55–56% is a good model, 57–58% is roughly where the betting market sits.

The line under each matchup

Tonight's announced starters, and what they are worth

The two names are the announced starting pitchers, pulled from the league's own schedule feed before first pitch. The number beside them is how far the starters move the home team's win probability, in percentage points, using each pitcher's prior-season FIP regressed toward league average. No result from this season touches it.

It is logged as a separate model rather than folded into the headline number. Measured against the full projection it gained 0.39% Brier on 2024 and 0.01% on 2025, which is not enough to overwrite a forecast that already has a track record. Both run side by side and the next few hundred games decide which is better.

Why some are marked provisional

An honest note about this slate

Predictions are written an hour before the earliest first pitch. When a slate is picked up late, any game already finished is excluded entirely — predicting a completed game is not a prediction — and any game already underway is marked provisional.

The model reads team strength and the schedule; it has no access to live game state, so a game being underway does not change the number. But "the model didn't peek" is a claim, and a timestamp that predates first pitch is proof. Only the clean ones are proof.

Benchmarks worth knowing

What good actually looks like in this sport

Baseball is the most random of the major sports. Over a full season the best team wins about 62% of its games and the worst about 38%, so even perfect knowledge of team quality caps single-game prediction near 57–60%.

53–54% is home-field advantage alone. 55–56% is a good model. 57–58% is roughly where the betting market sits. Anything claiming 70% on MLB games is measuring something other than wins and losses.