Win probability
The model's estimate of how often a team would win this matchup if it were replayed thousands of times under the same conditions.
Win probability is a single number between 0 and 100 percent that answers one question: if this exact matchup were replayed many times, how often would this team come out ahead? It is not a prediction of what will happen once. A 68 percent favourite is expected to lose roughly one game in three, and that is the model working correctly, not failing.
Pre-game win probability is built from opponent-adjusted efficiency, recent form, availability of key players, rest and travel, and sport-specific factors such as starting pitcher in MLB or expected goals profile in soccer. In-game win probability updates continuously as score, time remaining and possession change.
The only honest way to judge a win probability model is over a large sample. Group every 65 to 70 percent call together and check whether roughly 65 to 70 percent of them won. That check is called calibration, and AI4Gameday publishes the resulting record on its accuracy ledger.
Related terms
See it applied
Every published call on AI4Gameday carries a stated confidence and is graded in public once the game is final.