[warmly] Hey, it’s Duke Sizemore. It’s August 23, and this is one of those quiet NBA Sundays where the absence of movement is useful information. No fresh power mover cleared our threshold, there are no new league headlines in the feed, and the title-race prices we have are stale, so I’m not going to pretend they moved today. Instead, Boston gives us a really clean case study in why preseason Elo is not simply last season’s final number. And later, Detroit gives us a pretty interesting chemistry number and two October market disagreements worth putting side by side.
Quick model check-in first. The latest 25-game sample is still at 60 percent moneyline accuracy, with a 14.8-point average margin error. The broader tracked sample is much healthier: 1,246 games at 66.8 percent moneyline accuracy, with a .211 Brier score. Brier score is basically a test of probability quality: being confidently wrong hurts more than being cautiously wrong, so lower is better.
[thoughtful] The recent weakness still looks tied partly to injury handling. The model found that day-to-day tags with almost no estimated player impact were getting too much influence, especially while its news-correlation pipeline wasn’t providing the context it normally would. And the system has actually been tuning that setting: the day-to-day injury weight was reduced and then most recently moved from .14 up to .182 after the miss board suggested the earlier cut had gone too far for meaningful day-to-day absences. That’s the balance we’re trying to find: don’t move a projection because of a basically empty injury tag, but don’t ignore a questionable player who actually matters.
Now Boston. The Celtics had reached a 1716 Elo after beating Philadelphia 128-96 on April 26. Then they lost three straight to the Sixers: 113-97, 106-93 and 109-100. By May 2, that Elo was down to about 1669. The spotlight calculation has their recent swing at minus 38.9 points.
Elo here is just a rolling estimate of team strength. Beat an opponent, your rating generally rises; lose, it falls, with the size of the change depending on what the model expected. So a nearly 39-point swing isn’t saying Boston literally became 39 points worse on the scoreboard. It says the evidence from that run made the model materially less confident in Boston’s underlying strength.
[pause] And then preseason processing changes the question again. Boston’s carry Elo is 1610.2, while its market Elo is 1664.1. Blend those inputs and the preseason rating lands at 1637.1, fourth in this preseason context. In other words, we aren’t just photocopying the last rating from May. We’re combining what the team showed with what the market currently implies.
There is one concrete Celtics personnel item: Jordan Walsh agreed to a three-year, $15 million extension through 2029-30. Jayson Tatum is still listed day-to-day with a knee issue in the supplied injury feed. Given everything I just told you about stale and noisy injury tags, I would not turn that label alone into some grand October forecast.
Detroit is actually where today’s deeper player data gets fun. Elijah Harkless averaged 6.8 points, 2 rebounds and 2.9 assists over 26 games, and he currently has no health report. His on/off net figure in our chemistry sample is minus .506 across those same 26 games.
That on/off idea is pretty simple: compare how the team performs, in net-rating terms, while a player is on the floor versus when he’s off it. Harkless being at minus .506 means Detroit’s net performance was about half a point worse in the on-floor condition than the off-floor condition represented by this metric. That is not the same as saying Harkless personally costs Detroit half a point. Lineups, opponents and role all live inside that number, and 26 games is limited context. Among the Detroit names supplied today, Gary Harris is at minus .385 over 48 games and John Collins at minus .234 over 69, while Chaz Lanier is plus .12 over 33.
[curious] Actually, hold on, let me pull up Sonar real quick, because Harkless has a useful bit of background here. He played college ball at three different schools: Oregon, Oklahoma and UNLV. He became known more for defensive versatility than big scoring headlines, so that winding transfer path is part of the story of how he got here. That also helps explain why I wouldn’t evaluate him off 6.8 points per game and stop there.
The October board gives Detroit an immediate connection back to Boston. On October 20, Boston visits Detroit. The market has Detroit at about a 54.5 percent home probability and minus 1.5; our model has the Pistons at 65.2 percent and roughly minus 3.9. So that’s a 10.7-percentage-point probability disagreement, with our projected spread 2.4 points more Detroit-friendly than the market.
[genuinely surprised] The much bigger disagreement is October 21, Milwaukee at Washington. The market makes Washington about a 69.7 percent home favorite and has the Wizards minus 6.5. Our model goes the other way: only a 46 percent Washington win probability, with a model spread of Washington plus 1. That’s a seven-and-a-half-point spread gap. I’m not calling that a result two months early; I’m saying that is exactly the kind of game you circle and investigate because the market and model currently describe two very different teams.
There are a couple of real offseason personnel notes elsewhere. Dallas extended Naji Marshall for three years and $52.2 million. The Knicks have an executive change coming, with Gersson Rosas reportedly leaving after four years. And Victor Wembanyama is set to host some Spurs teammates for workouts in France.
Otherwise, quiet means quiet. No team cleared the power-mover threshold versus August 22, and the next confirmed game in the feed is Miami at Toronto in the preseason on October 3. There is no value in manufacturing a trend out of a blank board.
[chuckles] Sometimes the most data-scientist sentence imaginable is, “Nothing changed, and that matters.” But today it really does. Boston’s old slide is still a meaningful input, preseason Elo is deliberately pulling information from more than one place, and those October odds give us actual hypotheses to test once basketball starts answering back.
[warmly] That’s it for today. Thanks for spending a little of your Sunday with me. I’m Duke Sizemore, and I’ll catch you next time on COURT VISION: The Daily Brief.