Carleton Blew the Roof Off and the Model Had a Night

Three games, fifty-eight players graded, and the kind of slate where the injury news at tip-off ended up mattering more than almost anything drawn up on a whiteboard beforehand. Portland gave us the most interesting subplot — an injury reshuffled minutes in a way that handed real usage to players who might have been footnotes otherwise, and those players cashed. New York was supposed to be a two-headed story and ended up being mostly one head, very large, very productive. Washington gave us a rebounding clinic from someone who has been doing this quietly all season. And Golden State reminded everyone that the assist column is a real place to score DraftKings points if you know where to look.

The box scores, taken together, tell you something true about this league right now: the injury news cycle is fast, the role players who benefit from it are underpriced almost by default, and on nights when multiple bumps happen across multiple games, the variance in outcomes — for lineups and for projection models alike — gets wide in a hurry. Tonight was that kind of night.

The honest one-number receipt first: a mean absolute error of 6.3 DraftKings points across 58 graded players. That is not a number to brag about and not a number to apologize for — it is a number to look at clearly. Some of that error is random noise. Some of it has names. Let's go through them.

The Slate's Top Scorers

• Breanna Stewart, NY — We had her at 44.0. She finished at 47.5 on 32 minutes, 26 points, 8 rebounds, 3 assists, 1 steal, 3 blocks. An injury had bumped her minutes heading into tonight, and she rewarded every lineup that absorbed that information. A 3.5-point miss on a 47.5-point performance is the model doing its job.
• Bridget Carleton, POR — We had her at 28.7. She finished at 47.0. That is an 18.3-point miss, and it is the single biggest number in the brief, so it deserves honesty rather than spin. An injury had bumped her minutes, the model registered that and moved her up — she was in the best-call list — but the floor it gave her was still too conservative for what she actually did: 22 points, 8 rebounds, 4 assists, 3 steals in 32 minutes. At $8,300, she was the value of the slate by a considerable margin. More on her in a moment.
• Olivia Miles, MIN — We had her at 45.1. She finished at 43.5 on 34 minutes, 14 points, 4 rebounds, 10 assists, 4 blocks. The assist and block combination is unusual — that is the kind of line that is hard to project with precision — and a 1.6-point miss in either direction at this price point ($10,700) is the model earning its keep. An injury had bumped her minutes, too.
• Shakira Austin, WSH — We had her at 32.0. She finished at 41.5 on 31 minutes, 18 points, 16 rebounds, 1 assist, 1 block. Sixteen rebounds. At $9,700. The model did not fully anticipate the rebounding volume — nobody's does when a center goes for 16 — but a 9.5-point positive miss here is the kind of miss you can live with.
• Veronica Burton, GS — We had her at 31.3. She finished at 39.8 on 33 minutes, 11 points, 5 rebounds, 11 assists, 2 steals. An injury had bumped her minutes, and the assist total at $9,000 made her a genuine value play. An 8.5-point positive miss with the direction right is a clean result.
• Carla Leite, POR — We had her at 31.6. She finished at 39.5 on 31 minutes, 21 points, 4 rebounds, 7 assists, 2 steals. An injury had bumped her minutes. At $8,100, she was one of the cheaper paths to a 39-point DraftKings night on the board. A 7.9-point positive miss — direction correct, magnitude underestimated.

Best Calls

• Carleton is on this list because the model identified the minutes bump and moved her up from where she would have otherwise been priced and projected. The direction was right. The ceiling was higher than we said, but at $8,300 with a positive projection adjustment from an injury edge, she was flagged. That matters.
• Austin at $9,700 with a 32.0 projection that delivered 41.5 — the model liked the matchup and the role, and the rebounding volume did the rest. A correct call on a player who is not always in the conversation.
• Burton — correctly identified as a minutes-bumped value at $9,000, correctly projected to contribute across multiple categories. Eleven assists in 33 minutes is a genuinely hard thing to forecast, but the model had her in the right range for the right reasons.

Worst Calls

These are the misses. We own them.

• Kiki Iriafen, WSH — We had her at 30.1. She finished at 12.0 on 24 minutes, 6 points, 2 rebounds, 2 blocks. An 18.1-point miss in the wrong direction, at $9,600. The model did not flag a minutes bump here, but the actual production was a significant disappointment regardless of what drove it. This is the number that costs lineups the most on a night like tonight.
• Nia Coffey, MIN — We had her at 27.9. She finished at 14.8 on 20 minutes, 9 points, 3 rebounds, 1 assist. An injury had bumped her minutes in the projection, and either the bump did not materialize as fully as expected or the production did not follow the opportunity. A 13.1-point miss at $8,400 is painful.
• Sabrina Ionescu, NY — We had her at 34.8. She finished at 21.8 on 34 minutes, 12 points, 1 rebound, 4 assists, 1 steal. An injury had bumped her minutes projection, she played those minutes, and the production just was not there. A 13.0-point miss at $10,400 on a player who was supposed to complement Stewart — and instead faded into the background while Stewart did everything — is the kind of result that is genuinely hard to project. When Stewart absorbs that much of the offense, someone adjacent to her has to give something back. Tonight it was Ionescu.

A few things worth carrying into the next slate.

First: when an injury bumps multiple players on the same team — Portland tonight had Carleton and Leite both flagged — the model's job is to distribute that usage honestly, and then you still have to make a decision about which player is more likely to be the one who runs with it. Tonight both of them ran with it. That happens. When it does, the lineups that had the nerve to roster both walk away very happy.

Second: the Ionescu result is a real and recurring projection challenge. When a co-star on the same team has a ceiling game — Stewart at 47.5 — the surrounding players can give back usage in ways that are structurally hard to anticipate. The minutes projection can be right and the DraftKings output can still be wrong because the ball finds one set of hands. This is not an excuse; it is a feature of the game that the model needs to sharpen its read on when projecting co-stars at high price points.

Third: Shakira Austin's 16-rebound performance is a reminder that rebounding volume for a well-positioned center can spike in ways that turn a reasonable projection into a massive outcome. The model had the right player; it did not have the right ceiling. That is the kind of upside that GPP construction should be hunting — not because you can project 16 rebounds with confidence, but because the range on a player like Austin in the right matchup is genuinely wider than her salary suggests.

The model is walk-forward validated — it beats the naive baselines out of sample — and a 6.3 MAE on a high-variance, injury-shuffled three-game slate is honest. Some nights the edge is cleaner. Tonight it was messier, and the recap is the place to say that plainly. We will be back tomorrow with the next slate.