[warmly] Hey, it’s Duke Sizemore. It’s August 24, and today I want to spend some time in Memphis, because the Grizzlies are carrying one of the clearest negative signals in our preseason data. Not a rumor, not a one-day futures wiggle — actually, there wasn’t a qualifying market mover today — but a 58.4-point recent Elo downswing wrapped around a six-game losing streak. We’ll unpack why that still matters months later, check one place where our model and Vegas are miles apart for October, and I’ve got a Detroit player profile where the chemistry number needs a little context.
First, the model checkup. Over the latest 25-game sample, we’re at 60 percent moneyline accuracy with a 14.8-point average margin error. The larger tracked history is 1,246 games at 66.8 percent, so this recent window has clearly been weaker.
[thoughtful] And the useful part is that the model has identified a pretty specific problem. Day-to-day injury tags were influencing predictions even when the estimated player impact was basically zero, while the news-correlation pipeline that normally adds context had stopped producing tags. Detroit-Cleveland on May 18 is a rough example: Detroit was given a 75.2 percent home win probability and lost by 31, producing a 37.9-point margin miss, while Tristan Enaruna had a day-to-day tag with only a 0.1 impact estimate. New York also had huge wins while Jalen Brunson was tagged day-to-day with zero estimated impact.
The model has actually adjusted its own day-to-day injury setting in response. It cut the weight, then most recently nudged it from .14 to .182 because the earlier correction appeared to go too far when a day-to-day designation really did matter. That’s the balance we’re trying to find: don’t let a meaningless tag shove the prediction around, but don’t teach the model to ignore genuine availability uncertainty either.
Now Memphis. The Grizzlies finished 25 and 58, and their recorded schedule ended with six straight losses. They went from an Elo of 1333 before the April 4 Toronto game all the way down to 1274 after the April 13 loss in Houston. The ugliest single Elo hit came at Utah: Memphis lost 147 to 101 and dropped 24.6 Elo points from that game alone.
Elo, in plain English, is our running estimate of team strength. Beat a team you weren’t expected to beat and you gain more. Lose a game the model thought you should handle and you give more back. So a 58.4-point slide isn’t merely saying, “Memphis lost six games.” It says those results collectively forced a substantial downgrade in how strong the team looked.
[pause] The preseason model does give Memphis some regression back. Its final Elo was 1274.2, while its carry component is 1353.2. But the market component is only 1244, and blending those signals lands the Grizzlies at a 1298.6 preseason Elo, 29th in the league in this data. Their listed title probability is just 0.08 percent, with plus-100000 title odds.
There’s also a real roster wrinkle. Kentavious Caldwell-Pope has agreed to a buyout with Memphis and plans to join Philadelphia on a one-year deal after clearing waivers, according to the supplied ESPN report. And Memphis has several day-to-day designations in the feed, including Walter Clayton Jr., Zach Edey, GG Jackson and Ty Jerome. Given everything I just told you about the model’s injury-tag problem, I’m treating those labels carefully rather than turning every one into a preseason alarm bell.
Elsewhere, Dallas has extended Naji Marshall on a three-year, 52.2-million-dollar deal. Victor Wembanyama is set to host some Spurs teammates for workouts in France, and Chris Bosh has publicly cautioned Wembanyama regarding blood clots and taking his medicine. Those are actual offseason developments; we just don’t have a fresh futures-market response attached to them today.
Let me stay with Detroit for our player microscope, but focus on what the number means rather than just ranking a chemistry table. Elijah Harkless played 26 games in the supplied sample and averaged 6.8 points, 2 rebounds and 2.9 assists. He currently has no health report attached, so the data treats him as healthy.
His on/off net figure is minus .506. An on/off number compares the team’s net scoring performance — points scored minus points allowed — with that player on the court versus when he sits. Here, Detroit’s net performance was about half a point per 100 possessions worse with Harkless on the floor than off, in this sample. [curious] That’s mildly negative, not a verdict on a player. Twenty-six games is a limited sample, and on/off numbers absorb lineup quality, opponents and role. The interesting thing is that we can say precisely what the model sees without pretending the statistic tells us why by itself.
Actually, hold on, let me pull up Sonar real quick. [chuckles] This is a good Harkless footnote: Elijah Harkless is the younger brother of Shai Gilgeous-Alexander. They use different last names because Harkless uses his father’s surname while Shai uses his mother’s. That is exactly the sort of roster-page rabbit hole I enjoy more than I probably should.
And now the October board, because this is where things get spicy without us needing to manufacture any August drama. These games are well ahead of tip-off, so treat them as early disagreements, not finished predictions.
The largest probability disagreement in the posted set is Brooklyn-Charlotte on October 21. The market has Brooklyn at about a 41.3 percent home win probability and plus 4.5 on the spread. Our model has Brooklyn at only 16.4 percent and makes that spread roughly plus 10.1. So the model is much more favorable toward Charlotte: almost a 25-percentage-point gap in win probability and 5.6 points of spread disagreement.
[genuinely surprised] Washington-Milwaukee is nearly as interesting because the two sides actually disagree about who should be favored. The market makes Washington a 6.5-point home favorite on October 21, with roughly a 69.7 percent implied home probability. The model puts Washington at only 46 percent and has Milwaukee favored by about one. That is a seven-and-a-half-point spread gap. When a model and a sportsbook cross through zero like that, I pay attention — not because the model automatically wins the argument, but because it tells us exactly where our underlying assumptions are different.
There’s a big one the other way in New York on October 20. The market has the Knicks around 67.7 percent to beat Philadelphia and favored by 5.5. The model is much stronger: 90.2 percent Knicks, with a modeled spread of 13.8. That’s another game I want to revisit as availability and lineups become clearer, especially with Philadelphia now expected to add Caldwell-Pope.
One final housekeeping point: there were no qualifying power-rating movers from August 23 to August 24, and the title-race information is unchanged. The power-mover feed itself has also been frozen at zero movers since August 11 according to the model’s latest diagnostic, so I’m not going to interpret zero movement as proof that the market literally hasn’t moved. That feed needs a freshness check.
So Memphis is the signal today: a six-loss finish, a 58.4-point recent Elo drop, and a preseason blend that still has the Grizzlies sitting 29th. October is a long way away, but that gives us a clean baseline. Now we get to see whether health, roster changes and actual basketball can move them off it.
[warmly] Thanks for hanging out with me. I’m Duke Sizemore, and this is COURT VISION: The Daily Brief. I’ll catch you tomorrow.