Simulator · Monte Carlo
Pick a matchup and Mixl plays it out inning by inning — once, or fifty thousand times — on the same runs model that prices totals and run lines on the live board.
Every sim draws a full game: nine innings of runs for each team from the model's run distribution, extra innings if it's tied. Win probability is just the share of simulated games each side won — no formulas about averages, actual played-out outcomes.
Team runs are overdispersed — crooked innings happen — so the model samples runs from a negative binomial fit on decades of games: each team's rolling offense, the opponent's run prevention, the starter, the park, and the league run environment set tonight's expected runs.
Negative binomials add up exactly: nine per-inning draws sum to precisely the same distribution as the full-game model. The line score you watch isn't decoration — it's the pricing model, decomposed.
One sim tells one story. Noise shrinks with the square root of N: 100 sims pin win% to about ±10 points, 1,000 to ±3, 50,000 to ±0.4. That's why the pros run tens of thousands — drag the slider and watch.
100 sims → win% known to ±9.8 pts (95% band, worst case)