A probability with the working shown.

What the number means

Each percentage is the frequency of modeled winning margins. Soccer keeps zero-margin trials as draws. Other sports split zero margins evenly under an explicit overtime assumption. This is our simulation, based on the measured factors and game results described below.

How we calculate it

For each team, compare offensive factors with what its defense allows. Shrink small samples toward the league average, standardize across teams, and cap each factor at two standard deviations. Apply the displayed editorial weights. Fit the historical home-minus-away scoring margin to the difference in weighted factor ratings plus a home advantage term. The factor-rating slope is constrained to be nonnegative; home advantage is fitted separately and is zero at neutral venues. Each historical game uses only information available before that week or calendar date.

Bootstrap the calibration fits and sample observed margin errors to model possible outcomes. Game noise is scaled to the larger of fitted residual SD and chronological holdout margin RMSE. Stable seeds make each dated forecast reproducible. The factor effect chart shows the original fitted baseline, while simulated means include coefficient uncertainty.

Standard deviation

Margin SD describes the spread of simulated outcomes in points, runs or goals, depending on the sport. The 10th and 90th percentile margins show the middle 80 percent of simulations. Between-batch win-rate SD, in percentage points, describes sensitivity to bootstrap fits plus simulation noise. Monte Carlo standard error describes numerical sampling noise only. The outcome range comes from the simulated margin distribution.

NFL model

nflverse 2026 play-by-play, regular-season Weeks 1–4. Four games per team; league-average play/drive priors retained. Established NFL editorial weights 40/25/10/15/10 percent.

Completed source games 64; unavailable or incomplete source games 0; lagged calibration games 48; earlier training games 32; chronological held-out games 16. Holdout RMSE 9.07 points, Brier 0.254, winner accuracy 50.0%. Brier baseline 0.250. Fitted rating slope 0.922; home advantage 1.11 points; game-noise SD 12.47 points.

WNBA model

ESPN completed 2026 WNBA regular-season and playoff box scores, checked October 6. Five-game league-average prior; forecasts require five previous games for both teams. New provisional editorial weights 40/25/20/15 percent; not fitted factor coefficients.

Completed source games 343; unavailable or incomplete source games 0; lagged calibration games 302; earlier training games 214; chronological held-out games 88. Holdout RMSE 13.56 points, Brier 0.183, winner accuracy 72.7%. Brier baseline 0.250. Fitted rating slope 9.837; home advantage 2.45 points; game-noise SD 13.56 points.

COLLEGE-FOOTBALL model

ESPN 2026 college-football drive and play records, August 20–October 5. League-average play/drive priors; 50/70/100 percent success thresholds by down, 10-yard rushing and 20-yard passing explosive thresholds. Defensive rates are included. Turnovers use offensive drives ending in an interception or lost fumble; special-teams muffs are excluded. Football editorial weights 40/25/10/15/10 percent.

Completed source games 225; unavailable or incomplete source games 165; lagged calibration games 113; earlier training games 55; chronological held-out games 58. Holdout RMSE 20.78 points, Brier 0.224, winner accuracy 62.1%. Brier baseline 0.250. Fitted rating slope 12.486; home advantage 7.76 points; game-noise SD 20.78 points.

MLB model

ESPN MLB completed box scores and play records, August 15–October 5. Last 30 completed games with a five-game neutral net-factor prior. Power uses ISO; discipline uses (BB − K)/PA; running pressure uses (SB − 2 CS)/PA. Bullpen leverage uses relief traffic in inning seven or later within two runs: (2 × runs + baserunners)/PA, blended with whole-bullpen runs per inning. This is an editorial close-game pressure measure, not published leverage index. New MLB editorial weights 35/30/15/20 percent.

Completed source games 607; unavailable or incomplete source games 0; lagged calibration games 530; earlier training games 403; chronological held-out games 127. Holdout RMSE 4.07 runs, Brier 0.248, winner accuracy 54.3%. Brier baseline 0.250. Fitted rating slope 0.000; home advantage 0.16 runs; game-noise SD 4.43 runs.

SOCCER model

ESPN Premier League completed match statistics through October 5. Two-match neutral net-factor prior. Possession is measured directly. Corner share is a territory proxy. Opponent passes per tackle, interception or foul is a full-pitch pressing proxy. On-target shot share is a shot-quality proxy. The feed does not provide field tilt, zone-restricted PPDA or xG per shot; these proxies are labeled and are not those missing measures. Available soccer factor proxies weighted equally at 25 percent each.

Completed source games 50; unavailable or incomplete source games 0; lagged calibration games 40; earlier training games 30; chronological held-out games 10. Holdout RMSE 1.67 goals, Brier 0.645, winner accuracy 60.0%. Brier baseline 0.667. Fitted rating slope 0.706; home advantage -0.09 goals; game-noise SD 1.67 goals.

Coverage and limits

Published forecasts cover all confirmed NFL, college-football, MLB, WNBA and Premier League games in the seven-day window. College football uses success, explosive-play rate, starting field position, finishing drives and turnovers; baseball uses power, discipline, run pressure and measured close-game bullpen pressure. NBA regular-season games are outside this publication’s seven-day window. Preseason is excluded.

Ratings do not adjust for opponent strength, injuries, availability, starting pitchers, rest or travel. The inputs, observation dates and chronological checks are shown above.