Where the Model Gets It Wrong: The NHL's Most Unpredictable Teams in 2026

The model correctly picks Vancouver games 70% of the time. It gets Islanders games right 38.5% of the time. A coin flip would literally do better on Islanders games.

That's a 31.5 percentage point gap between the most and least predictable teams this season. Across 1,086 games, the spread between teams tells a story the top-line number hides.

Some teams are clockwork. Some are chaos. Understanding the difference matters more than any single prediction.

#The predictable

These teams have made the model's job easy.

 

The pattern is obvious. Teams at the extremes are easy to predict.

Vancouver is bad. A -75 goal differential means they're getting outscored by more than a goal per game. The model just picks the other team and is right 7 out of 10 times. Colorado and Dallas are the mirror image. +79 and +51 goal differentials. They win at home, win on the road, and don't randomly collapse against bad opponents. Consistency is what models reward.

#The chaos teams

Then there's the other end.

 

Pick accuracy is how often the model called the winner correctly. Expected vs. actual win percentage shows how far off the model's view of each team was from reality. Three teams where the model is wrong more often than it's right, each for a different reason.

#Buffalo: a GM firing changed everything

I wrote about the Sabres' playoff push a couple weeks ago. What I didn't get into was how badly the model whiffed on them.

The model predicted Buffalo would win 53.6% of their games. They've actually won 80%. That's a 26.4 percentage point gap, the widest calibration miss in the league.

On December 15, the Sabres fired GM Kevyn Adams and promoted Jarmo Kekalainen. At the time, Buffalo was 14-14-4 and sitting last in the Atlantic. Here's what happened after:

 

A franchise-record-tying 10-game winning streak in December. Then another 8-game streak after that. Rasmus Dahlin is putting up Hart Trophy numbers with 60 points in 61 games. They have 10 players with 30+ points; no other team has more than 8.

The model sees stats. It doesn't see a front office shakeup that changes the entire culture overnight. By the time the on-ice numbers caught up to the new reality, the model had already gotten dozens of games wrong. This is the clearest example of something no model can do: predict inflection points driven by human decisions off the ice.

#Florida: the season that started with a torn ACL

On September 25, the very first day of training camp, captain Aleksander Barkov collided with defenseman Niko Mikkola during practice. Torn ACL. Torn MCL. Surgery the next day. Out 7-9 months.

That's your two-time defending Stanley Cup champion's best player, their Selke-caliber center, gone before they played a single game.

 

The model predicted Florida would win 56.5% of their games. They've won 40.6%. The model adjusts for injuries, but it underestimated how much Florida's system depends on Barkov specifically. Without him anchoring lines and the penalty kill, the team's identity changed. Some nights they'd gut out a win. Other nights they'd collapse. The model kept expecting them to play like the Panthers. They weren't the Panthers anymore.

#The Islanders: 10-0 in overtime and nobody knows why

The Islanders are the model's worst team at 38.5%, and unlike Buffalo or Florida, there's no clean explanation.

They're 38-24-5 with 81 points. Solid record. But here's the thing: the Islanders are the first team in NHL history to go 10-0 in overtime. Those 10 wins generated 20 extra standings points. Last season they lost 9 of 14 overtime games. This year, perfect.

 

Ilya Sorokin has 7 shutouts, tying the franchise record and leading the league. But on nights he doesn't shut the door, games are tight enough that they often go to overtime. Then the Islanders win. Every time.

No model can reliably predict a team going 10-0 in overtime. That's roughly a 1-in-1,000 probability if each OT is a coin flip. The model predicted a 54.5% win rate; they're at 64.3%. The gap is almost entirely overtime magic.

That's the honest answer. The model doesn't get the Islanders, and I don't think any model would.

#What this tells you

The model is great at confirming what's already obvious. Bad teams lose, great teams win, and the extremes are predictable. Where it struggles is with regime changes, injury cascades, and teams that operate in the margins.

None of this means the model is broken. A+ picks still land at 79.3%. But understanding where the model is confident and where it's guessing is the difference between using it well and using it blindly.

The teams at the top and bottom of this list are a map of the model's blind spots. Knowing the blind spots is half the value.

---

Check today's predictions to see confidence grades for every game. For more on how the model thinks about luck and regression, read What is PDO in Hockey?

Accuracy figures from 1,086 games tracked October 9 through March 25, 2026. Updated daily at /performance.