cfb_model() loads a model as an xgboost booster and errors if its feature names no longer match cfb_model_manifest.
Initial release: the college football models of the cfb_model_artifacts release (expected points, win probability with and without the spread, overtime reach, field goal, fourth down conversion, two point conversion, expected pass, completion probability with and without air yards, and quarterback rating) as raw xgboost vectors, plus the punt distance distribution, the field position expected points table, the model cards and the bundle manifest. Bundle version 2026.10.12, the models sportsdataverse-py scores with as of that date.
Bundle 2026.10.12 against 2026.09.09:
cfb_qbr_model is the 2026-09-30 retrain on the features the play-by-play processor serves. It drops the spread feature (9 features, not 10), so a consumer that builds a fixed column list must take the feature names from the booster (xgboost::getinfo(m, "feature_name")).
New cfb_cp_air_yards_model, the completion probability model for passes with known air yards, and cfb_wp_ot_reach_model, the probability the game reaches overtime.
The manifest is now generated from the bundle files by data-raw/MANIFEST.R; the 2026.09.09 card hashes predated the card upload and are corrected.
Hex logo: an isometric battery cell whose lid shows six decision trees wired into the terminal (gradient boosting), drawn by data-raw/hex-logo/.
