Two families of skater metrics side-by-side. ΔV is a decision-quality proxy: for every on-puck event, we score V(after) − V(before) from the pretrained state-value network, sign-adjusted to the actor's team. Positive ΔV = the actor's team's expected-goal outlook improved over the next 1.5 s. xG is a shooting-selection + finish proxy: total expected goals from shot geometry (XGBoost, leave-one-game-out CV), actual goals, and the finish delta G − xG.