resolve wcbb
D1 women's basketball — Bayesian per-player projections from a hierarchical NB2 with per-(position × experience) quadratic. Same architecture as the WNBA pipeline, fit independently on the women's college substrate (328k player-games × 7.6k players × 17.3k games). Walk-forward 2025 OOS accuracy 70.4% · log-loss 0.574 (vs always-home baseline 0.608, +9.6pp).
players
4,541 players · per-36 PTS/REB/AST/STL/BLK/TOV with 90% intervals · sortable, searchable, filterable by team or position. Top of the board (PTS/36): Ta'Niya Latson, Audi Crooks, Ayoka Lee, JuJu Watkins, Paige Bueckers.
teams
358 qualified D1 teams · projected offense, empirical defense, net rating, strength_z. UConn #1 (net rating +31.0), Notre Dame, Baylor, USC, South Carolina round out the top 5.
matchup predictor
Pick any two teams · get a predicted margin and win probability. Normal-margin model with HFA = 4.16, slope = 0.86, σ = 14.46 — fitted on 5,180 games. Toggle neutral-site to drop the home advantage.
methodology
3-level hierarchical NB2: league → position → player with non-centered parameterization. Per-position quadratic on experience (Fr/So/Jr/Sr) since DOB isn't available in any NCAA source. No per-player exp tilt (NCAA's 1.7-season window won't identify it).
how the model works
What's in the model: 3-level hierarchy league → position → player, non-centered on position and player offsets. Per-position quadratic on experience with peak ~ N(2.5, 0.7) (Jr/Sr peak) and gamma_pos ~ Exp(2) curvature. NB2 likelihood, vectorized. Per-(athlete, season) z-scored context features (pace, team quality, mates' usage, gravity) ready as Phase 4 substrate — fold into Stan in v2.
What's not in v1: no per-player experience tilt (sigma_exp_player not identifiable with the 1.7-season average career window — same call as WNBA + the men's pipeline). No per-row recency weighting yet — deferred to v2 along with the live in-season blend.
Validation: walk-forward 2025 OOS: accuracy 0.704, log-loss 0.574, vs always-home baseline 0.608. +9.6pp lift over the prior. HFA = 4.16, MARGIN_SIGMA = 14.46 (wider than NBA's 12.0; expected for a less-stratified talent distribution).