NHL careers on the Hall's doorstep

Method and calibration

One logistic regression for the NHL, fitted on 2,846 retired, eligible players of whom 218 are in the Hockey Hall of Fame (base rate 8%). Retired, last season ≤ 2023 (at least one induction class chosen since he became eligible), and ≥ 5 seasons in the VAL corpus. Inductees below the floor are excluded from training. 10 inductees fall below that floor — mostly careers played largely before or outside this league's corpus — and are not trained on. League-best seasons are ranked within VAL group (e.g. batters vs pitchers, skaters vs goalies).

The model reads on-field value only: career and peak VAL, league-best seasons by VAL, playoff VAL and longevity. It cannot see awards voting, championships beyond playoff VAL, pre-league or international careers, eligibility bans, or character clauses. Every training player's probability is out-of-fold (scored by a model that never saw his own label). Active players (appeared in one of the last two seasons — the corpus has no retirement flag) and retired players still inside the waiting period are scored on their career to date by the full-data model — a trajectory, not a forecast. A non-inductee who is still eligible may yet be chosen, and a player still playing outside the league is not eligible however long ago he left it. Black ink is shown for context but is not a model input: its season coverage is shorter than the VAL corpus.