Frame-by-frame puck and skater positions from 10 hockey games (league not disclosed by the data release), turned into four things you can rank players on: how good their shot chances are (xG), how much their decisions swing the game (ΔV), how well they position themselves without the puck (ΔPC), and how their choices compare to what an AI "coach" would pick. Every number is clickable back to the play it came from.
Every page reads static JSON/CSV from web/data/. No backend.
Scrub through any goal / shot / zone entry frame-by-frame. Puck & skaters, jerseys, model values.
Open viewer →Actor-perspective value delta per on-puck action, aggregated per player.
Open leaderboard →Per-player top speed, distance skated, high-speed time %, acceleration bursts from smoothed velocities.
Open skating →Counterfactual pitch-control contribution per skater when they’re off the puck. The MARL companion to ΔV.
Open off-puck →Every shot with distance, angle, shot type, and cross-validated xG probability.
Open shots →Training curves and holdout metrics for V(s), BC, Q(s,a), CQL. Policy action distributions.
Open dashboards →