PlayDecoded

Model Accuracy

Every percentage on this page is outright accuracy: did the team we favored win. For how that differs from Brier score and calibration, see the accuracy glossary.

Factor model scores every current game. Each model's record is shown separately, never blended into one number.

Weekly report cards

Each week lists every finished game, the pre-kickoff win probability, and this week's hit rate and Brier score next to the season-to-date numbers on this page.

Performance by sport and season

Calibration this season

Calibration asks whether our probabilities match reality: teams we gave about 70% should win about 70% of the time. This season's finished games are grouped by the pre-game probability we assigned the leading team; each point is that group's average predicted probability versus its actual win rate. We call a band "Well calibrated" when its actual rate is within 5 percentage points of the predicted value. Bands with fewer than 10 games are marked low-sample because a handful of results is mostly noise.

Live factor model, MLB, 2026. Finished games only; preseason excluded. Same grading rule as the rest of this page.

Predicted versus actual win rate for MLBScatter of average predicted probability against actual win rate by probability band, with a diagonal line for perfect calibration. 1856 finished games.50-55: predicted 54.4%, actual 51.9%, 1379 games55-60: predicted 56.1%, actual 54.4%, 399 games60-65: predicted 62.5%, actual 67.6%, 34 games65-70: predicted 66.5%, actual 42.9%, 14 games70-75: predicted 72.5%, actual 73.7%, 19 games75-80: predicted 75.8%, actual 62.5%, 8 games, low sample80-85: predicted 80.4%, actual 100.0%, 3 games, low sample50%50%70%70%90%90%Predicted probabilityActual win ratePerfect calibration
Calibration by probability band for MLB. Average predicted probability versus actual win rate. Bands with fewer than 10 games are marked low-sample.
Probability bandGamesAverage predictedActual win rate
50-55137954.4%51.9%
55-6039956.1%54.4%
60-653462.5%67.6%
65-701466.5%42.9%
70-751972.5%73.7%
75-80875.8%62.5%
80-85380.4%100.0%
85+0——

Calibration by Confidence

When the model says 70%, does the favored team actually win ~70% of the time? Points above the dashed line outperform; below underperform.

50-55%: predicted ~53%, actual 51.9% (1379 games)55-60%: predicted ~57%, actual 54.4% (399 games)60-65%: predicted ~62%, actual 67.6% (34 games)65-70%: predicted ~67%, actual 42.9% (14 games)70-75%: predicted ~73%, actual 73.7% (19 games)75%+: predicted ~78%, actual 72.7% (11 games)50%75%100%50%75%100%Perfect calibration
Model ConfidenceGamesCorrectActual Accuracy
50-55%137971651.9%
55-60%39921754.4%
60-65%342367.6%
65-70%14642.9%
70-75%191473.7%
75%+11872.7%

Week-over-Week Trend

How the model has performed over recent weeks. The dotted line marks 50% (coin flip).

50%
W0
W14
W17
W20
W23
W26

How We Measure Accuracy

A prediction is correct when the team our model gives > 50% win probability actually wins the game. Ties are excluded from accuracy calculations.

Calibration measures whether our confidence levels are meaningful. If we say a team has a 70% chance of winning, they should win roughly 70% of those games -- not 90%, not 50%.

For context, a coin flip gives 50% accuracy. Our model targets outright winners: predicting which team wins the game.