In-game win probability
Win probability for every game in progress, recomputed as the score and inning change. The model is fit from our own inning-by-inning data rather than borrowed from a published table.
And what it deliberately leaves out
The estimate is P(home team wins) given the run differential, the inning and which half is being played, fit from every half-inning state in 600 completed games.
Base-out state is not included yet. Runners on second and third with one out genuinely matters, but it moves win probability by a few points while score and innings remaining move it by tens. Adding it is a real improvement and is listed as future work rather than quietly implied.
One check worth mentioning: the first fit was not monotone — the raw sample claimed that being two runs down in the third was better than being one run down. That is impossible, it was small-sample noise, and a sanity check caught it before this page existed. Isotonic regression fixed it.