The Lab · Leaderboards
2026-27As of Wed, Sep 30

Whole-league leaderboards

Rank every qualified skater on any of 30 metrics — the on-off impact, puck-security, and finishing-luck numbers most sites never surface. Pick a metric, pick a pool, and read the whole league sorted the model's way. Percentiles are always position-relative.

How to read a leaderboard

A leaderboard answers one question: out of everyone who plays this position, who is best at this one thing?

The colored bar is the percentile, with a tick at the league median. Filled far to the right is good. The word beside it — ELITE, STRONG, AVERAGE — is the same information for readers who would rather not do arithmetic.

Some metrics are better low (goals against, giveaways). The board already sorts each one the right way round, so the top of the list is always the good end. Full glossary →

Pctile
Where a player sits out of 100 comparable players. 90th percentile means only a tenth of his peers do it better.
Position-relative
Defencemen are ranked against defencemen, forwards against forwards. A blue-liner is never graded on a winger's scale.
Qualified
A minimum-ice-time cut, so a player with two good shifts cannot top a list. The pool size (n = …) is printed on every board.
*
An asterisk marks a model-estimated number — our best reconstruction of something the public data does not publish, never an official stat.

On-Ice HD xGA / 60

n = 0 qualified D

High-danger expected goals allowed per 60 with this player on the ice at 5-on-5 (lower is better defensive suppression).

5on5 · lower is better — best on top (list inverted) · xGA/60

#TeamPlayerValuevs. position

◆ Method · Single-metric leaderboard over players with games_played >= minGP in the metric's situation; percentiles are position-group relative. Data: NHL stats API + Backcheck xG (data/mp_skaters_2025.csv). · percentiles are position-relative, model-estimated proxies from NHL data and our own xG model via the analytics sidecar.