How BaseballPicksHub predictions work
A plain-language guide to our KBO, NPB and MLB win-probability model, calibration, and how to read the numbers.
We publish free win-probability predictions and full statistical breakdowns for every KBO, NPB and MLB game. This page answers the questions people search for most about baseball prediction models.
Why you can trust these numbers
Our model is not a static guess. It learns from real outcomes every day, and every prediction is tracked in the open.
Learns from real outcomes
Every game gets compared to its actual result. The model recalibrates itself on that feedback, and new versions only go live after passing an independent performance check on held-out games.
Calibrated to real frequencies
When we say 60%, those games really do win about 60% of the time over a large sample. A calibration layer keeps our stated confidence aligned with what actually happens.
Track record in the open
Every prediction is logged at publish time and graded publicly. You can see each game on the results page and the rolling accuracy and Brier score on the track-record page.
We never hide the misses. Every game stays in the archive so you can judge the model on its full record — not on cherry-picked wins.
How the model works
For every game we combine four measured signals — starting-pitcher matchup (ERA, FIP, WHIP, K/9, days rest), team offense (wRC+, runs per game, recent form), bullpen freshness (last 14 days ERA and IP), and park & weather (park factor, wind, temperature, rain risk) — then adjust for market and tipster consensus. The output is a single win probability per team plus a projected run total, refreshed multiple times per day.
FAQ
How accurate are your baseball predictions?
We track every published prediction with a Brier score (lower is better; below 0.25 beats a coin flip) and a calibration table that compares predicted win rates to the actual win rates in each probability bucket. All numbers are public on the Results page, updated after every game finishes.
What is a Brier score and why does it matter?
The Brier score measures how far a probabilistic prediction is from what actually happened, squared. A perfect forecaster scores 0. Naïve 50/50 guessing scores 0.25. Any number below 0.25 means the model beats a coin flip on average — the standard sanity check for win-probability models.
Do you cover KBO, NPB and MLB?
Yes. We publish predictions for every scheduled KBO (Korea), NPB (Japan) and MLB (USA) game. The homepage shows all games starting in the next 24 hours, and you can filter by league.
Where do the odds come from?
Market odds are scraped from public bookmaker aggregators when available. Odds and edge values are only shown to visitors in regions where display is legal — everywhere else we show pure model probabilities.
How is the win probability calculated?
We simulate expected runs for each team using a Poisson model calibrated on starter, bullpen, offense (wRC+), recent form, park factor and weather. Head-to-head history is blended in when available. The simulated run distribution gives both a win probability and a projected total.
What does 'Why X%?' mean on a game card?
It breaks the prediction into share-of-signal contributions: how much of the model's confidence comes from offense, from the starting pitcher matchup, from bullpen state, and from park & weather. Factors can interact, so shares won't always add up to exactly 100%.
How often is the data updated?
The scraper runs multiple times per day. Pitcher assignments, bullpen usage, weather and market odds are refreshed on a rolling window. Finished games are graded within about 15 minutes of the final out.
Is this a betting tip service?
No. BaseballPicksHub is a free statistics and prediction site. We don't sell picks, run a tipster subscription, or take wagers. The numbers are published so you can make your own informed decisions.
What is a 'Strong pick' badge?
A game earns Strong pick when three things line up at once: high model confidence, a real edge versus market odds, and multiple independent factors (offense, pitcher, bullpen, park) all favoring the same side.
How do I read the projected score?
The projected score shows each team's expected runs from the model, blended with recent head-to-head history when meaningful sample exists. It's a mean — not a prediction of the exact final score — so most games land within a couple of runs on either side.