The Hockey Hall of Fame announced its class this month, and the part that always bothers me is the people it didn't announce. A committee meets behind a closed door, argues for an afternoon, and reads out a list. No vote totals, no runners-up, no math. You learn who got in. You never learn who finished a hair short, or by how much, or why.

So, I built a model that has to show the math — trained on every Hall of Fame induction in NHL history, handing every player who ever laced them up a single number: the probability they belong in the Toronto-based museum. This month I got to check it against a live verdict. It passed — and then it did something more interesting than pass.

It Called the Class — Now Stop Trusting That

Frozen before this year's vote, the model ranked the four NHL men the committee chose — Keith Tkachuk, Carey Price, Patrice Bergeron, Pekka Rinne — at 20th, 27th, 33rd, and 170th out of 3,064 eligible players. Three of the four sit in the top one percent of everyone who has ever been eligible. It had Bergeron — twenty seasons, a Cup, an NHL-record six Selkes, in on the first ballot — as a flat lock, and Price, the goalie some still call a compiler, flagged as overdue before the committee ever got to him.

That is the easy part to brag about and the wrong part to trust. Any model can be tuned to flatter a class after you know the answer. The honest test runs the other way: freeze it in the past and make it guess forward. So I ran it on every Hall class since 2010 — the Hall inducts annually; 2021 was the lone skip, after the pandemic shoved the 2020 ceremony back a year. Each year the model saw only the players already enshrined, ranked the board, and I checked where that class's eventual inductees actually landed.

The median inductee sat in the top 0.6 percent of a board the model built without being told who would get in. Sixty percent landed in the top one percent; nearly nine in ten in the top five. Freeze it in 2018 and it has Martin St. Louis and Martin Brodeur ranked one and two on a board 2,600 deep, years before the committee made it official. Sixteen classes, the same answer every time. Not a fluke, and not hindsight.

The Misses Are the Argument

The proof that the thing understands hockey isn't in its hits. It's in its misses. When the model whiffed — when an inductee ranked nowhere near the top — it was almost always the same kind of name. Sergei Makarov, 160th. Rogie Vachon, 174th. Mike Vernon, 463rd. Václav Nedomanský, at 1100. Those aren't random errors. They're the veteran-committee and character picks the hockey internet relitigates every June — short NHL careers, defectors whose prime happened overseas, role players honored for what they meant rather than what they scored.

The model, reading only an on-ice NHL résumé, shrugs and says it doesn't see it. The argument around those names says the same thing. It agrees with the committee exactly where everyone agrees and disagrees exactly where everyone fights. That's what you want from a referee, not a fan.

Connor McDavid's Six Ghosts

The model doesn't really score a player. It compares him. Underneath the probability is the question you'd ask at a bar — who does this guy remind you of? — and the answer is a list of the most similar careers in history. Feed it Connor McDavid: 794 games, 1,220 points, three Harts, six Art Rosses. The six players his career most resembles come back Phil Esposito, Guy Lafleur, Stan Mikita, Bobby Hull, Mario Lemieux, and Jean Béliveau. Six for six, every one in the Hall, half of them on hockey's shortest list. He isn't chasing the Hall of Fame; he's chasing Lemieux's ghost, and at 29 years old he's most of the way there.

It works as a character read, too. Bergeron's six Selkes make him the best defensive forward who ever lived, and the model's nearest comps for him come back Pavel Datsyuk, Bob Gainey, Jere Lehtinen — the exact names you'd reach for if someone asked what kind of Hall of Famer he was. That isn't luck: the model carries a specific bump for the rare forward who wins the Selke again and again, so it didn't just rate Bergeron a lock — it rated him a lock for the right reason.

Carey Price and the Compiler Label

Goalies are where Halls of Fame go to argue, and the model's favorite disagreement this year was Carey Price. He took Montreal within two wins of a Cup in 2021 and never got closer; he won the Hart, the Vezina, and the Ted Lindsay in a single staggering 2015 and never won any of them again. To a certain kind of fan, one monster year and one near-miss adds up to a compiler.

The model never bought it. It rated Price a clear Hall netminder, 27th and overdue, and the company it kept tells you why — Rinne, Miikka Kiprusoff, and Henrik Lundqvist, or the workhorse generation that dragged ordinary teams into the playoffs on their backs. Strip away the win-loss record a goalie barely controls, look at what he actually stopped, and Price stops with the best of his era. The gap between what a goalie is blamed for and what he did is exactly the kind of thing a model sees more clearly than a memory does.

Who's Next, and What It Would Cost

The model doesn't stop at the retired. Project the active stars forward and it sorts them into tiers. A handful are already locks in waiting — Alex Ovechkin, Patrick Kane, McDavid, Nathan MacKinnon, Nikita Kucherov, Leon Draisaitl, Evgeni Malkin, Sidney Crosby, and Cale Makar, the only defenseman it files with the forwards who cannot miss. A step back sit the near-certainties: Anze Kopitar, Connor Hellebuyck, Auston Matthews, and Andrei Vasilevskiy. Then the real fun, the borderline tier, where a career is a live argument right now: Mikko Rantanen, Kris Letang, Mitch Marner, Artemi Panarin, Claude Giroux, Brent Burns, Quinn Hughes, Matthew Tkachuk, Roman Josi.

And because the model re-scores a whole hypothetical career, you can ask it the bar question directly: what single thing moves a man up a tier? One more Vezina turns Sergei Bobrovsky or Vasilevskiy from likely into lock. A Conn Smythe does it for Erik Karlsson. A first Hart or Art Ross lifts David Pastrnak or Panarin off the bubble. And the blue line question of the era lands in one line — Letang and Burns have the longevity, the points, the playoff pedigree, but the model, like a lot of voters, wants another Norris for Burns, or any Norris for Letang, before it calls either one a lock.

One Thing It Can't Do

A model like this answers in or out, and at that it's excellent. It is not built to rank legends against each other, and if you push it to, it'll cheerfully tell you Mark Messier outranks Wayne Gretzky — because at the very top everyone pins to the ceiling and the tiebreakers collapse into volume and Cups. Read it the way it's meant to be read: tiers and trajectories, not a greatest-of-all-time list.

The committee tells you who got in. The model tells you who belongs, who's a hair short, and which ghost the next one is chasing — and on the only question they both answer, it has quietly been right for 16 years running.

How It Works, Briefly

Strip the jargon and the model is a pattern-matcher. I gave a machine the full career of every player ever eligible for the Hall — scoring, peak rate, trophies, playoff production, international play, longevity — and told it which ones got in. It learned the shape of a Hall of Fame career the way you learn a face: not one rule but a thousand small ones weighted against each other. For the curious, it's a gradient-boosted decision-tree classifier, calibrated so its output reads as an honest probability.

Two pieces sit on top. A similarity engine finds each player's nearest historical comparables — the McDavid-to-Lemieux part — by measuring distance across all those numbers at once. And two hand-built fixes cover what the raw model reads poorly: goalies, whose win totals lie about them, get a separate résumé index, and repeat-Selke forwards like Bergeron get a credit the box score misses. Crucially, it never grades its own homework — every accuracy figure here comes from freezing the model before a vote and making it rank blind.

Sources

Career and season statistics: the NHL's official stats API, covering every player and season back to 1917, re-pulled and audited against the league's own published totals.

Hall of Fame inductions: the Hockey Hall of Fame's player record, cross-checked name-by-name against Wikipedia's *List of members of the Hockey Hall of Fame*.

Awards and honors: Hart, Vezina, Norris, Selke, Conn Smythe, Art Ross and All-Star selections, Stanley Cups, and international appearances, assembled from NHL award records.

All figures reflect careers through the 2025–26 season.