10,000 Parallel Universes: What Monte Carlo Actually Means
We don't predict one outcome. We simulate ten thousand of them and show you the distribution.
The Problem With a Single Prediction
Imagine someone asks you: "Who wins tonight, Dodgers or Padres?" You could say "Dodgers" and be done with it. But that throws away almost all the useful information.
How do the Dodgers win? By one run? By six? Does the Padres' bullpen collapse in the 7th? Is it a 2-1 pitchers' duel?
A single prediction is a point on a map. A Monte Carlo simulation is the entire terrain.
How It Works
For every game on the schedule, we take the actual starting lineups and run the game through our Markov chain model 10,000 times. Each simulation:
- Plays through all 9 innings (plus extras if tied)
- Uses real batter-vs-pitcher transition probabilities
- Randomly samples from those probabilities at each at-bat
- Records the final score
Same lineups, same probabilities, but different random outcomes each time — just like real baseball. The randomness isn't a flaw. It's the point.
What 10,000 Games Look Like
Here's a realistic example of what the simulation output looks like for a single game — say, Dodgers vs. Padres:
Dodgers vs. Padres — 10,000 Simulations
But the score distribution is where it gets interesting:
Most Common Outcomes
In 3 out of 10,000 simulations, the Padres win 13-0. That's not a bug — it's a reflection of reality. Blowouts happen. Shutouts happen. The simulation captures the full spectrum because baseball is wild like that.
Why 10,000?
Why not 100? Or a million?
100 sims
±5%
Too noisy
10,000 sims
±0.5%
Sweet spot
1,000,000 sims
±0.05%
Overkill, too slow
At 10,000 simulations, our win probability estimates are stable to within about half a percentage point. Running more wouldn't meaningfully change the predictions — it would just take longer. We need to generate predictions for every game, every day, so speed matters.
The margin of error (±0.5%) is far smaller than the inherent uncertainty of the game itself. The noise is in baseball, not in our sample size.
What We Extract From the Distribution
Those 10,000 simulated games give us far more than just "who wins":
Win probability: Dodgers won 5,700 of 10,000 sims = 57.0% win probability
Expected total runs: Median total across all sims — say, 7.5
Score distributions: Full histogram of home and away scores
Blowout probability: What percentage of sims had a 5+ run margin?
Extra innings probability: How often did it go past 9?
A single prediction says "Dodgers win." A Monte Carlo simulation says "Dodgers win 57% of the time, the most likely score is 4-3, there's a 12% chance of extra innings, and in 3% of simulations someone scores 10+ runs."
The Name
Monte Carlo methods are named after the famous casino in Monaco — not because they're about gambling, but because they use randomness as a computational tool. The technique was developed during the Manhattan Project by Stanislaw Ulam, who was playing solitaire and realized he could estimate probabilities through repeated random trials faster than calculating them analytically.
The same idea powers everything from nuclear physics simulations to financial risk modeling to weather forecasting. And, in our case, simulating whether the Dodgers' cleanup hitter goes yard in the 5th inning across 10,000 parallel timelines.
Uncertainty Is the Feature
When you see our prediction say "57% Dodgers," that 57% isn't hedging. It's the actual, honest probability derived from simulating the game ten thousand times with the best data we have.
The Padres' 43% isn't noise to be ignored — it's 4,300 fully simulated universes where San Diego's bats got hot, or LA's starter didn't have it, or the sequencing just broke their way.
Every game is 10,000 parallel universes. We just tell you which ones show up most often.