Model Accuracy

Backtested on 7,289 MLB games across 2023–2025. Every game simulated 10,000 times.

56.1%

Win Prediction

57–62%

O/U at 9.5

64–69%

O/U at 10.5

3.39–3.65

Score MAE

Win Prediction by Season

Markov, Elo, and ensemble blend accuracy across three full seasons.

Win Probability Calibration

When we say a team has X% chance of winning, how often do they actually win? Perfect calibration follows the dashed line.

Over/Under Accuracy by Line

Accuracy of over/under calls at different total run lines, averaged across all three seasons.

Score Prediction Error

Mean Absolute Error for total score predictions (lower is better). Our median-based predictions outperform mean-based.

Full Season Breakdown

Season Games Markov Elo Blend O/U 9.5 O/U 10.5 Score MAE
2023 2,430 53.8% 56.1% 56.5% 56.8% 64.4% 3.65
2024 2,429 54.4% 56.2% 56.2% 61.7% 68.5% 3.39
2025 2,430 52.6% 56.7% 55.6% 60.4% 67.3% 3.61

About these numbers

The theoretical accuracy ceiling for MLB game prediction is probably 58–62%. Baseball is inherently random — the best team in baseball loses 40% of its games. Our ensemble model approaches that ceiling by combining player-level Markov chain simulations with team-level Elo ratings. All results are out-of-sample: the model never sees future data when making predictions.