[warmly] Hey, it’s Duke. I’m glad you’re here.
We are officially in that quiet pocket of the calendar — no games until October 3rd, Miami at Toronto — but honestly? This is one of my favorite times of year. Because when the ball stops bouncing, the numbers get loud. And right now, the loudest number on the board belongs to the New York Knicks.
Today I want to talk about that 41.5-point Elo surge they rode into the offseason… why it matters, why it mostly doesn’t carry over… and what the model is actually seeing when it says New York is in the mix, but not quite in the OKC–San Antonio tier.
And we’ll check in on how the model’s been behaving lately — including a pretty honest look at why we keep getting smoked by 30-point blowouts. [chuckles] Those have been… humbling.
So let’s start with how we’ve been doing.
Last season the model tracked 1,246 games. Moneyline accuracy: 66.8%. The average favorite probability was 68.3%, and favorites won 66.8% of the time. That’s what calibration looks like — when your 68% calls win about 68% of the time. Brier score at 0.211, which basically measures how far our probabilities are from reality — lower is better — and that number sat right in line with market-grade models.
Now, zoom in to the most recent sample — last 25 picks — we’re at 60% accuracy. Not disastrous, but not crisp either. And the thing that jumps out isn’t direction. It’s margin.
Our average margin error is 14.8 points.
[pause]
That’s not “we were a possession off.” That’s “the game script detonated.”
The latest self-evaluation the model ran is titled — and I love that it names its own problems — “Injury tags aren’t explaining the blowout misses.”
Here’s what that means in plain English. We looked at the five biggest margin misses. Most of them were games where the favorite still won… but won by 25 or 30 more than we expected. New York at Cleveland — plus 35.4 versus expectation. San Antonio at Minnesota — plus 26.7. New York at Philly — plus 26.7.
We were right about who was better. We were wrong about how violently better.
And when we checked injuries, the listed absences were mostly negligible in our impact table — zero or near-zero net rating effect. So it’s not like we missed a 7-point superstar downgrade. The issue is confidence calibration. We’re pricing too many games in that 60–75% “competitive favorite” range… when some of these late-season matchups had real 30-point blowout volatility baked in.
The one true faceplant? Detroit hosting Cleveland. We gave Detroit a 75.2% win probability. Cleveland won by 31.
[genuinely surprised] Thirty-one.
That’s not an injury miss. That’s volatility. That’s a reminder that team-level ratings sometimes understate how wide the distribution can get when one side collapses.
And that brings us to the Knicks.
Because if you want a case study in late-season volatility — the good kind — it’s New York.
They close 2025–26 at 70–32. Final Elo: 1808. That’s elite. Over their final stretch, they added 41.5 Elo points.
Quick refresher on Elo, by the way. It’s not a power ranking vote. It’s a probability engine. Every team starts with a number. When you win, you take points from the opponent. How many points depends on expectation. Beat a strong team? Big gain. Lose to a weak team? Big drop. Blowout margins nudge it further.
So when I say New York gained 21.6 Elo points in a 130–93 win at Cleveland on May 26… that’s because they beat a good team, on the road, by 37. That’s the system saying, “Okay, something real just happened.”
Over their last six tracked games, four wins over San Antonio. The final one — June 14, 94–90 on the road — added another 7.5 Elo.
And here’s the part I can’t stop staring at.
Across their last four tracked games: 128.8 offensive rating. 102.6 defensive rating. Plus-26.2 net rating. And they did it at a 96.7 pace — deliberate, controlled, half-court heavy.
[thoughtful] That’s dominance without chaos. That’s surgical.
So naturally you’d think… okay, preseason title favorite, right?
Not quite.
Because preseason Elo resets.
Their carryover from 1808 drops to about 1701.1. The market Elo is 1702. That’s the model blending last season’s strength with regression toward league mean and uncertainty about roster continuity, health, offseason noise.
That’s why their title probability sits at 7.71%, +950 odds.
Very good.
Not top tier.
At the top, it’s still Oklahoma City and San Antonio. Thunder at 24.55% title probability, 1751 Elo. Spurs right behind at 21.89%, 1750 Elo. That’s razor thin — essentially coin-flip separation in rating space.
And it’s quiet. No futures movement since August 5th. No new power movers. This is continuity, not momentum.
Which is important. Because in the offseason, one day of odds wobble can look like a trend. But right now? The board is stable.
Now, there’s an elephant hovering over New York’s preseason optimism.
Jalen Brunson.
He’s technically listed day-to-day. But he’s set to undergo surgery on his right wrist/forearm area and is expected to need at least two months of rehab.
[pause]
That’s not nothing.
In the injury impact table we use for margin adjustments, Brunson was graded at zero in those blowout samples — which tells you something about how the model views marginal day-to-day tags versus systemic role value. But two months is structural. If he misses meaningful early-season time, you’ll see that preseason Elo get tested immediately.
And this is where I’ll say something that might sound contradictory: the model doesn’t panic in August. It waits for games.
Elo moves on results. Not vibes.
Speaking of vibes… actually hold on, let me pull up Sonar real quick. [curious]
Okay, this is fun.
Top riser in hype metrics right now is Philadelphia — finished 50–44 last year. And Sonar reminds me: the franchise began as the Syracuse Nationals in 1946. The name “76ers” was chosen to honor the signing of the Declaration of Independence in 1776 after a rename contest.
[laughs softly] I love that. Imagine crowd-sourcing a name and landing on something that historically on-brand.
And since we’re in trivia mode — one more.
Let me check the Clippers note. [pause]
Yeah — the Clippers are one of the few major North American pro teams to survive three different city identities. Buffalo Braves. Then San Diego Clippers in ’78 — named after the sailing ships. Then Los Angeles in ’84, and they kept “Clippers.”
That’s such a strange continuity arc. Same name. Three cities. Same franchise heartbeat.
And right now? They’re actually our biggest hype faller — minus 109.7 in that measure after a 42–41 finish. Which tells you perception cooled fast.
Now, quick zoom back to Boston.
Jayson Tatum was ruled out for a Game 7 against Philadelphia with a knee issue. That’s the kind of razor-thin margin moment that swings seasons. Boston’s preseason Elo sits at 1637. Title probability 5.4%. Still in the tier, but clearly behind the top three.
This is what I mean when I say August numbers are anchored in last year’s signal. New York’s 70–32 matters. That +26 net rating pocket matters. OKC and San Antonio’s sustained strength matters.
But the model is also whispering: volatility lives here.
Especially in that 60–75% favorite band.
So here’s a practical takeaway for when games actually start October 3rd — Miami at Toronto.
Early season probabilities in the high-60s? Treat them as confidence with humility. We are actively skeptical of moderate favorites in volatile profile matchups. That’s straight from the latest learning note.
No upcoming games yet beyond that confirmed opener. No Summer League standings — zero games logged there. This is pure calibration week.
And honestly? I kind of love that.
Because when there’s no noise, you can really see the structure.
Right now the structure says:
— OKC and San Antonio are co-leads in the title tier.
— New York’s late heater was real, but preseason regression pulls them into a strong third slot at 7.7%.
— Boston and Detroit trail behind in the next band.
— The futures market hasn’t budged in 24 hours.
— And our biggest vulnerability as a model is underestimating how lopsided some “competitive” games can become.
[thoughtful]
That last one matters to me. Because it’s easy to celebrate a 66.8% season and move on. It’s harder — and more interesting — to ask why the misses looked the way they did.
Were we anchoring too tightly to base ratings? Probably.
Were we assuming late-season intensity compresses margins? Looks like it.
Did reality remind us that sometimes a team just hits the gas and the other side runs out of road? Absolutely.
That’s not failure. That’s information.
And information is the whole point.
So as we sit here in early August, staring at a quiet board, I’m honestly excited. Because the first two weeks of real games are going to tell us whether New York’s 41.5-point surge was a springboard… or just the cleanest four-game stretch of the year.
Either way, the model will adjust. Elo always does.
[warmly] Alright. That’s your coffee chat for today.
If you like living in this space between numbers and story — between probability and personality — I’m glad you’re here with me.
I’m Duke Sizemore. This is COURT VISION: The Daily Brief.
And we’ll talk tomorrow.