first 20 games tell

Why Your First 20 Games Tell You More Than You Think — And Less Than You Fear 

Twenty games into the 2025-26 season, a simple points-pace projection had Carolina finishing with 114.8 points. They finished with 113. At the same point, the same projection had Buffalo finishing with 74. They finished with 109, a 35-point miss.

Same sample size, same method, wildly different results. Understanding why those two numbers diverge is more useful than either number on its own.

The conventional wisdom, and why it's incomplete

The standard line in hockey is that early standings are noise: don't overreact to a hot or cold start, and wait for the sample to grow. That is not bad advice, it's just incomplete. Unfortunately early standings can give decision-makers false confidence in doing nothing.

We tested it two ways. First, we projected final points from the first 20 games for 16 teams in 2025-26. The average miss was about 10 points. That's close enough to separate a playoff team from a lottery team, but not close enough to settle a bubble race.

Second, we ran a wider test across 156 team-seasons, covering every franchise over five seasons where full game logs exist. Our test asks how accurately the early win pace predicts final wins, and how that accuracy changes as games accumulate. And we discoverd that there is no magic threshold. Error falls steeply through the first 15 games, and after that each additional five games buys roughly half the improvement of the five before. By game 20 the typical miss is about six wins. By game 30 it's about four.

So 20 games isn't a switch that flips from noise to signal. It's the point where the early record becomes worth taking seriously, if you know what to check.

The more useful finding is what determines whether your team's 20-game projection belongs with Carolina or with Buffalo.

What separates a Carolina from a Buffalo

Carolina's projection worked because what Carolina did in its first 20 games was what Carolina always does. Their CF% (Corsi For percentage, the share of all 5-on-5 shot attempts a team generates) averaged 58.9% over five seasons. That's the highest in the league and more than four points ahead of the second-best team in CF%, which is Florida. Carolina's CF% through 20 games in 2025-26 was 60.2%, right in line with the 59–60% they had posted in each of the prior three seasons. When a team's early underlying numbers match its established identity, 20 games tend to look a lot like 82.

Buffalo is the more interesting case. Buffalo's possession numbers weren't unstable. Their season-to-season CF% has actually varied less than Carolina's over five years. Through 20 games they were being out-attempted, at 47.5% CF%, and losing. That's a coherent picture of a below-average team, and the projection read it accurately.

What changed after game 20 wasn't Buffalo's possession, which finished the season at 49.3%, still below break-even. What changed was their finishing and goaltending. Buffalo's full-season PDO was the second-highest in the league. Their record outran their shot share. That's why the miss was so large: no early-season model built on possession or pace would have predicted what happened next.

Utah shows the opposite failure mode. After a hot start, their 10-game projection was a wildly optimistic 131 points. By games 16–20 it had settled into the 92–96 range, within a few points of their actual 92. Their CF% through 20 games, 53.2%, matched their prior season exactly. The hot start was noise, and the underlying numbers said so all along. The way to catch a distorted small sample isn't to ignore the data. It's to check what's driving it.

PDO: a lens, not a verdict

PDO is a team's shooting percentage plus its save percentage, often expressed on a scale where league average is near 1000. A team running well above average is scoring and saving at a rate its shot volume doesn't fully explain. Historically, that tends to drift back toward the middle.

The conventional logic says an extreme PDO in the first 20 games should make you distrust the record. Our 2025-26 sample complicates that logic. In a small sample of 16 teams, the clubs with extreme PDO at game 20 were projected more accurately on average, not less. There was a modest lean in the expected direction: Colorado (1027) and Anaheim (1019) both ran hot early and finished well short of their 20-game pace. But Edmonton and Dallas carried extreme readings and landed within three points.

Buffalo is the bigger lesson. Their PDO through 20 games was unremarkable. Their surge came later. PDO isn't something to check once at game 20 and file away. It's worth tracking all season, because a sustained move in either direction is often the first sign that the record is about to separate from the underlying play.

Three questions at game 20

If you're an analytics staffer trying to give a straight answer to a GM or coach who wants to know whether the first 20 games mean anything, here's a usable starting framework:

1. Is our CF% consistent with our recent history? If it's in line with the last two or three seasons, as Carolina's and Utah's were, the sample is more trustworthy.

2. Where is our PDO relative to league average, and where is it heading? An extreme reading is a reason to look closer. A sustained shift after game 20 is a reason to re-run the projection.

3. Does our points pace match our underlying shot metrics? Out-attempting opponents and winning is aligned signal. Winning while being out-attempted means something less durable is propping up the record.

If all three line up, the projection is a reasonable baseline. If they diverge, the divergence itself is the finding. It's telling you to investigate before anyone makes a personnel or coaching decision on 20 games of standings alone.

It is worth noting that this framework would not have predicted Buffalo. At game 20, all three checks described a team playing to its level. Some seasons turn on what happens after the sample is taken, which is why the right move at game 20 is a baseline and a plan to re-check at games 30 and 40, not a final verdict.

The broader point

Analytics tools earn their keep when they go beyond what everyone can already see on the scoreboard to identify the gap between what the record shows and what's happening underneath it. Twenty games is neither too early to know anything nor late enough to know everything. It's enough data to know which questions to ask next, and when to ask them again.

© 2026 ORRO AI Genius, LLC. All rights reserved.