NHL Fantasy Draft Kit

What these numbers are, exactly

Each line is games x ice time x scoring rate, projected separately over up to the last three completed seasons with empirical-Bayes shrinkage toward position and frozen age curves, on an 84-game schedule. Validated walk-forward across 10 past seasons: mean absolute error 9.237 points, against 10.311 for the best simple alternative.

PPP, SOG, HIT and BLK are not projections. The season model projects games, ice time, goals and assists; it does not project shots, hits, blocks or power-play points. Each of those four is the player's own 2025-26 per-game rate carried forward and multiplied by his projected 2026-27 games -- it holds his role fixed, which is an assumption, not a forecast. A player with no 2025-26 rate has no value in these categories and no category score at all; he shows a dash, never a zero.

Category values are z-scores standardised within the draftable pool -- the top 250 skaters by projected points and the top 40 goalies by projected wins. Players outside the pool are scored against the same mean and standard deviation, so they rank (negatively) without defining the scale. Skater and goalie values are computed in separate pools and are not comparable to each other.

Tiers are gaps, not grades. Within a position, walking the board best-first, a drop of more than 0.5 standard deviations between one player and the next starts a new tier (at most 6). A tier break says the players above it are separated from the players below it by more than the usual step; it does not say the two groups are different in kind. The board is rebuilt by hand when the projections change or rosters settle, not nightly — a 2026-27 draft board does not move between Tuesday and Wednesday, and a nightly timestamp on a file that did not change would only make it look fresher than it is. Players with no NHL history are absent. There is no row to project and no way to know who makes a roster.