The licensed CFBD Model Training Pack: 4,520 games, 86 columns, 72 usable features, 2016–2024 (Week 5 onward). Feature importances below are read from the pack's own fitted artifacts — not re-trained here.
Ridge predicts Home Margin in points. Positive favors home. This saved-fit comparison does not change primary recommendations or estimate win probabilities.
Ridge comparison unavailable for 2026 Week 9. frozen environment mismatch
| Metric (points) | Result | 95% block interval |
|---|---|---|
| MAE | 12.46 | 12.01 to 13.07 |
| RMSE | 15.58 | 14.90 to 16.33 |
| BIAS | -0.64 | -1.88 to 0.51 |
Previously inspected season; source publication times unknown. Original training-target defect retained. Not prospective or betting evidence. Intervals use 13 season/type/week blocks; they do not guarantee future performance.
Historical head-to-head accuracy is unavailable without a verified matched primary forecast artifact. These are Ridge-only errors.
Retained history, not current predictions. Generated Oct 29, 2026, 06:22:30 CDT; cutoff Week 8.
Kent State at Sacramento State: Home Margin -2.42 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin -2.42 points.
Southern Miss at UL Monroe: Home Margin -5.36 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin -5.36 points.
Notre Dame at Navy: Home Margin -27.52 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin -27.52 points.
South Carolina at Oklahoma: Home Margin +3.70 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +3.70 points.
Oregon State at Fresno State: Home Margin +1.15 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +1.15 points.
Florida Atlantic at North Texas: Home Margin +19.82 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +19.82 points.
Michigan at Rutgers: Home Margin -15.43 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin -15.43 points.
Baylor at UCF: Home Margin -3.01 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin -3.01 points.
Texas State at Boise State: Home Margin +12.20 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +12.20 points.
Florida at Georgia: Home Margin +17.17 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +17.17 points.
Virginia at Wake Forest: Home Margin +12.80 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +12.80 points.
Tulane at Charlotte: Home Margin -16.58 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin -16.58 points.
Arizona at Texas Tech: Home Margin +3.16 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +3.16 points.
Oklahoma State at Iowa State: Home Margin +1.72 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +1.72 points.
Clemson at Florida State: Home Margin +5.46 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +5.46 points.
Stanford at Louisville: Home Margin +25.33 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +25.33 points.
Illinois at Maryland: Home Margin -6.47 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin -6.47 points.
Nevada at UCLA: Home Margin +17.52 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +17.52 points.
Temple at East Carolina: Home Margin +5.10 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +5.10 points.
Army at Memphis: Home Margin +9.46 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +9.46 points.
New Mexico at San José State: Home Margin -10.96 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin -10.96 points.
Kansas at TCU: Home Margin +2.86 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +2.86 points.
California at NC State: Home Margin +10.26 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +10.26 points.
Washington State at San Diego State: Home Margin -3.72 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin -3.72 points.
Missouri at Arkansas: Home Margin -8.27 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin -8.27 points.
Arizona State at BYU: Home Margin +14.21 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +14.21 points.
UAB at South Florida: Home Margin +10.51 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +10.51 points.
Coastal Carolina at Georgia State: Home Margin +6.57 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +6.57 points.
Northern Illinois at UNLV: Home Margin +19.54 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +19.54 points.
Florida International at Liberty: Home Margin +17.43 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +17.43 points.
Marshall at Old Dominion: Home Margin -5.62 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin -5.62 points.
Purdue at Penn State: Home Margin +16.28 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +16.28 points.
Bowling Green at Western Michigan: Home Margin +11.70 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +11.70 points.
SMU at Syracuse: Home Margin -12.35 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin -12.35 points.
Boston College at Duke: Home Margin +7.65 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +7.65 points.
Wisconsin at Iowa: Home Margin +2.39 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +2.39 points.
Minnesota at Indiana: Home Margin +28.33 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +28.33 points.
Colorado State at Utah State: Home Margin +7.60 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +7.60 points.
UTEP at North Dakota State: Home Margin +27.23 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +27.23 points.
App State at Georgia Southern: Home Margin +5.47 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +5.47 points.
Georgia Tech at Pittsburgh: Home Margin +10.05 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +10.05 points.
Miami at North Carolina: Home Margin -18.42 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin -18.42 points.
Washington at Nebraska: Home Margin +7.73 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +7.73 points.
UConn at Air Force: Home Margin -2.34 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin -2.34 points.
Louisiana Tech at South Alabama: Home Margin -3.59 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin -3.59 points.
Ohio State at USC: Home Margin -9.41 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin -9.41 points.
Northwestern at Oregon: Home Margin +4.26 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +4.26 points.
Auburn at Ole Miss: Home Margin +12.67 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +12.67 points.
Kansas State at Colorado: Home Margin -8.58 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin -8.58 points.
Utah at Cincinnati: Home Margin -11.24 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin -11.24 points.
Mississippi State at Texas: Home Margin +15.33 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +15.33 points.
Troy at James Madison: Home Margin +21.64 points
Training-only baseline +2.85 points. Positive contributions favor the home team.
Baseline + contributions = Home Margin +21.64 points.
Paired RF, standalone XGBoost, corrected FastAI, standalone Logistic and the tutorial stack remain unavailable under their separate holds/evaluation gates. SHAP is an explanation workflow, not another predictive model.
Ridge explanations use independent-feature linear SHAP with all 3,916 pre-2024 training rows as background. They explain the exact evaluated Ridge fit, not causal effects or the separate historical XGBoost margin model below.
Model SHA256: 835eaeeb0c3f24dc70d607cba4db93562ba0791fe0cc5f683e899cba588c9c37
Input SHA256: unavailable
Environment SHA256: unavailable
Games
4,520
Columns
86
Usable features
72
Seasons
2016–2024 (Week 5 onward)
Opponent-adjusted, no future leakage — each row's stats use games prior to that matchup only.
The spread card, displayed win probabilities and training examples have different jobs. Weekly research results do not replace a production model. These roles describe the existing code; each forecast view reports its own availability.
Produces the in-season spread card and matchup calculator. Its margin-derived probabilities are separate from the StackedEnsemble.
Target: Home Margin (home points minus away points)
src/cfb_model_lab/betting.py; ADR-0011
No reviewed standalone artifact catalogued.
External comparison and the announced preseason/early-season fallback where supported. It is not a locally trained model or an ensemble probability.
Target: External point ratings and rating-derived margins
src/cfb_model_lab/picks.py; CONTEXT.md (Engine)
No reviewed standalone artifact catalogued.
Notebook 01 fits Ridge with alpha=1.0. The supplied fitted artifact supplies the coefficient chart below; it does not produce the production spread card.
Target: Margin (away points minus home points)
model_pack/01_linear_regression_margin.ipynb
model_pack/ridge_model.joblib
Two RandomForestRegressors demonstrate team-score prediction. These are distinct from the RandomForestClassifier inside the production-code ensemble. No standalone score artifact is supplied here.
Target: Home points and away points; implied score margin
model_pack/02_random_forest_team_points.ipynb
No reviewed standalone artifact catalogued.
The supplied standalone classifier supplies the importance chart below. A separately fitted XGBClassifier is also an ensemble component; this artifact does not itself produce the spread card.
Target: Home-win probability
model_pack/03_xgboost_win_probability.ipynb; src/cfb_model_lab/model.py
model_pack/xgb_home_win_model.pkl
The legacy artifact is not a probability model. Probability output is unavailable from it. The corrected classification notebook trains separate private research models.
Target: Unbounded home-win regression score
model_pack/04_fastai_win_probability.ipynb; model_pack/legacy-fastai-requirements.txt
model_pack/fastai_home_win_model.pkl
Standalone interpretable classifier tutorial. Logistic regression is separately fitted as an ensemble component and meta-model; no standalone notebook artifact is supplied here.
Target: Home-win probability
model_pack/05_logistic_regression_win_probability.ipynb; src/cfb_model_lab/model.py
No reviewed standalone artifact catalogued.
Logistic regression, RandomForestClassifier and XGBoost feed a logistic stacker. Its displayed win probabilities do not select the EPA spread card. Expanded weekly refits remain research candidates until the promotion gate and owner's decision.
Target: Home-win probability
src/cfb_model_lab/model.py; scripts/build_ensemble_predictions.py; ADR-0011
public/api/ensemble.json
Reported local experiment, not integrated or reproduced by this catalog. No fitted artifact, performance result or production role is claimed.
Target: Research margin experiment
Owner-reported local research; no reviewed artifact in this checkout
No reviewed standalone artifact catalogued.
Coefficients predict training Margin (away points minus home points). Negative Margin favors the home team, opposite the dashboard’s Home Margin. Each coefficient is the change per feature unit; raw magnitudes are not directly comparable across differently scaled features. Faded bars are negative coefficients.
Gain-based importance. Spread alone drives a third of the model.
Original supplied artifact verified with legacy FastAI 2.7.19. It produces unbounded regression scores trained on home_win, not probabilities. The corrected classification tutorial produces separate private research models. Importances not available.