[warmly] Hey, it’s Duke Sizemore. I’m really glad you’re here. Late July basketball is this quiet little pocket of the calendar where nothing is technically happening… and somehow everything is happening. Today I want to talk about Philadelphia quietly stacking smart veterans around LeBron, Oklahoma City making a marginal move that actually matters at the top of the title board, and Washington trying to climb out of one of the steepest Elo slides we tracked all of last season.
And we’ll do our usual thing — check the model’s pulse, talk about what it’s getting right, what it’s still learning, and why some of these preseason numbers look the way they do.
[pause] Okay. First, the self-check.
Across 1,246 tracked games last season, COURT VISION finished with a 0.668 moneyline accuracy and a 0.211 Brier score. If you’re newer here, moneyline accuracy is straightforward — did we pick the winner? The Brier score is about calibration. It rewards us for saying “this team has a 70% chance” when that team actually wins about 70% of the time. Lower is better. So 0.211, paired with that 0.668 hit rate, tells us we weren’t just picking favorites blindly — we were reasonably aligned with reality.
Here’s the part I like: favorites carried an average 0.683 win probability in our forecasts… and they actually won at a 0.668 clip. That’s tight. That’s what calibration is supposed to look like.
Now zoom in to the most recent 25-game evaluation sample. We’re at 0.600 moneyline accuracy with a 14.8-point average margin error. [thoughtful] That margin error is where the story lives. Five games blew up — 24 to almost 38 points off. And the model did something I genuinely appreciate: it wrote its own little audit explaining that the problem wasn’t the injury weight settings… it was that the injury table assigned zero or trivial impact to certain absences.
In plain English: if the database says a player is “Day-To-Day” but gives them a 0.0 impact score, the model treats that like background noise. So the Elo rating — the core team strength number — carries full weight. And when that “minor” absence is actually meaningful? You get torched.
Elo, quick refresher. It’s a rating system originally built for chess. Every team has a number. When you beat someone good, you gain more points. When you lose to someone weaker, you drop more. It’s constantly adjusting based on opponent strength and margin. Think of it as a rolling résumé score.
So when the résumé says Team A is much stronger, and the injury input says “don’t worry about that guy,” the forecast leans hard into Team A. That’s how you end up with a 75% home win probability that flips into a 30-point loss.
The encouraging part? In the one game where the injury table actually registered a real, non-zero impact — the forecast nudged correctly and the favorite won comfortably. That tells us the weights themselves are directionally sound. The data feeding them needs cleanup. That’s a very fixable problem.
[curious] And I love that we’re not pretending otherwise. Misses are information. They’re not something to hide — they’re breadcrumbs.
Alright. Let’s talk actual teams.
Philadelphia is the clearest roster upgrade on the board this week. Kentavious Caldwell-Pope agreed to a buyout with Memphis and plans to sign a one-year deal with the 76ers after clearing waivers. That comes just days after LeBron James signed a two-year, $8 million deal with Philly.
[chuckles] That sentence still sounds surreal when I say it calmly.
From a modeling standpoint, this is fascinating because preseason Elo is heavily anchored in prior ratings. We’re still in “carryover mode.” There isn’t fresh game data yet to reshape team strength — so roster moves are the big signals.
KCP matters because he’s a clean fit. High-level perimeter defender, reliable spacer, low-usage, playoff-tested. In lineup math terms, he’s the kind of player who tends to improve on/off net ratings quietly. On/off net rating, if you haven’t heard me explain it before, is simply the team’s scoring margin per 100 possessions with a player on the floor versus off. It’s chemistry in numeric form. Not vibes — possessions.
When you pair LeBron’s playmaking gravity with credible shooting and defense on the wing, you stabilize lineups. And stabilization is gold in preseason modeling because volatility is what kills forecasts.
Meanwhile at the very top of the title race, San Antonio sits at a 1,752 Elo. Oklahoma City is right behind at 1,747. Both carry identical 0.2284 title probabilities. Literally tied at 22.84%.
That’s razor-thin separation.
The Thunder getting Spencer Jones on a two-year, $12 million offer sheet — with Denver holding the option to match — isn’t a blockbuster. But at that tier, marginal gains matter. Five Elo points can swing home-court probabilities in a playoff series. Two tenths of a percent in title equity is real when you’re already above 20%.
[pause] It’s like compound interest. You don’t feel it in July. You feel it in May.
Now let’s flip to the other end of the emotional spectrum.
Washington.
The Wizards closed last season 17–65 with a league-worst 1,201 Elo. Over their final stretch they dropped 48.3 Elo points. That’s not a slump. That’s a slide.
In April alone they lost six straight. Three successive home defeats dropped them -23.0, -9.7, and -7.8 Elo points. Those are significant hits because home losses to comparable teams get punished harder in this system.
Here’s the interesting part: preseason modeling bumps Washington up to a 1,340.5 Elo. That ranks 25th — up five spots from where they finished.
[genuinely surprised] That’s not nothing.
Why the jump? Carryover adjustments plus market inputs. The preseason Elo blends last year’s finish with regression toward league average and offseason signals. It’s basically saying: even bad teams aren’t usually as bad as their worst six weeks.
Their title probability is still 0.0033 — about a third of one percent. So we’re not talking contender. We’re talking stabilization.
Roster-wise, they’ve got Anthony Davis listed day-to-day. No. 1 overall pick AJ Dybantsa was shut down for the final three Summer League games — rest, precautionary. Summer League itself is now complete — 76 games in the books. Golden State finished 5–1. San Antonio and Orlando both 4–1. Fun samples, but remember: Summer League performance barely touches preseason Elo unless it reveals something structural about depth or usage.
And speaking of preseason… there are no confirmed upcoming games yet beyond October 5th — Lakers at Kings. So right now, this is all structural modeling. No fresh win probabilities to preview. Just architecture.
[thoughtful] Which honestly is my favorite phase. It’s like tuning an instrument before the concert.
One more macro note before we land this plane.
Indiana is flagged as our top “hype” riser in preseason modeling at +96.5, despite finishing 19–63 last year. The Clippers, on the other side, show a -114.3 hype drop after going 42–41.
That hype metric isn’t Elo. It’s sentiment-plus-movement — how far market and projection signals shifted relative to last year’s baseline. It doesn’t guarantee performance. It just tells you where expectations moved.
And expectations, in the NBA, are half the psychological game.
So here’s where we are.
Philadelphia made one of the cleanest veteran upgrades available. Oklahoma City is playing the margins at the top. Washington is trying to turn a -48.3 slide into a controlled rebuild climb. The model is calibrated well overall, recently dinged by injury-impact zeros, and deliberately not overcorrecting until the injury table itself improves.
That’s the key lesson today: don’t twist the knobs if the sensor is broken. Fix the sensor.
[laughs softly] That might be the nerdiest sentence I’ve said all month.
But it matters. Because when October 5th rolls around and Lakers-Kings tips off, every percentage point you see from this model will be built on this quiet calibration work happening right now.
I love this part of the season. It’s honest. No noise yet. Just numbers, context, and a little bit of patience.
[warmly] Thanks for spending a few minutes with me. We’ll keep watching the roster edges, keep stress-testing the injury inputs, and keep treating every miss as a clue.
I’m Duke Sizemore. This is Court Vision. And I’ll talk to you tomorrow.