On the moneyline the model's win probabilities scored a log loss of 0.6056 against the de-vigged closing price's 0.6058 — a dead heat. The one place it showed anything is where a moneyline is staked here, at 10%+ EV — nearly always an underdog the closing spread rated better than its price did: +15.9% on 782 bets (±8.8), positive in 5 of 7 seasons. Suggestive, not proven: a bucket needs 2.23σ to survive the looks taken, and the prices behind it are one archive's closing numbers. Closing-line value will settle it. Every game is still predicted, graded and CLV-tracked; the numbers below are worth tracking rather than backing.
Season
Every game is predicted and graded whether or not it is staked. Break-even at −110 is 52.38%; a moneyline record is shown beside the rate its prices implied, because a favourite winning 70% of the time is what the price already said.
The board
Where these numbers come from
Our model starts from the line and adjusts it for what team ratings, injuries, rest and travel say the line missed. How far it moves is what a pick is based on.
The percentages below use a blend: the line moved 40% of the way toward our model on the spread (POR -6.9) and 5% on the total (232.7). That much and no more, because over seven seasons the closing line has been the more accurate number.
Ratings only, ignoring the line entirely: POR -4.4 · 244.1. The model's independent opinion — and the least accurate of the three.
Where these numbers come from
Our model starts from the line and adjusts it for what team ratings, injuries, rest and travel say the line missed. How far it moves is what a pick is based on.
The percentages below use a blend: the line moved 40% of the way toward our model on the spread (CLE -9.3) and 5% on the total (235.3). That much and no more, because over seven seasons the closing line has been the more accurate number.
Ratings only, ignoring the line entirely: CLE -4.6 · 229.3. The model's independent opinion — and the least accurate of the three.
Where these numbers come from
Our model starts from the line and adjusts it for what team ratings, injuries, rest and travel say the line missed. How far it moves is what a pick is based on.
The percentages below use a blend: the line moved 40% of the way toward our model on the spread (HOU -2.7) and 5% on the total (219.6). That much and no more, because over seven seasons the closing line has been the more accurate number.
Ratings only, ignoring the line entirely: HOU -3.3 · 226.1. The model's independent opinion — and the least accurate of the three.
Where these numbers come from
Our model starts from the line and adjusts it for what team ratings, injuries, rest and travel say the line missed. How far it moves is what a pick is based on.
The percentages below use a blend: the line moved 40% of the way toward our model on the spread (ATL -3.8) and 5% on the total (240.6). That much and no more, because over seven seasons the closing line has been the more accurate number.
Ratings only, ignoring the line entirely: NOP -4.2 · 243.2. The model's independent opinion — and the least accurate of the three.
Where these numbers come from
Our model starts from the line and adjusts it for what team ratings, injuries, rest and travel say the line missed. How far it moves is what a pick is based on.
The percentages below use a blend: the line moved 40% of the way toward our model on the spread (MEM -8.5) and 5% on the total (230.6). That much and no more, because over seven seasons the closing line has been the more accurate number.
Ratings only, ignoring the line entirely: MEM -10.0 · 238.9. The model's independent opinion — and the least accurate of the three.
Where these numbers come from
Our model starts from the line and adjusts it for what team ratings, injuries, rest and travel say the line missed. How far it moves is what a pick is based on.
The percentages below use a blend: the line moved 40% of the way toward our model on the spread (SAS -12.1) and 5% on the total (228.5). That much and no more, because over seven seasons the closing line has been the more accurate number.
Ratings only, ignoring the line entirely: SAS -8.2 · 225.9. The model's independent opinion — and the least accurate of the three.
Where these numbers come from
Our model starts from the line and adjusts it for what team ratings, injuries, rest and travel say the line missed. How far it moves is what a pick is based on.
The percentages below use a blend: the line moved 40% of the way toward our model on the spread (PHI -1.5) and 5% on the total (212.5). That much and no more, because over seven seasons the closing line has been the more accurate number.
Ratings only, ignoring the line entirely: CHA -4.2 · 214.6. The model's independent opinion — and the least accurate of the three.
Where these numbers come from
Our model starts from the line and adjusts it for what team ratings, injuries, rest and travel say the line missed. How far it moves is what a pick is based on.
The percentages below use a blend: the line moved 40% of the way toward our model on the spread (BOS -4.9) and 5% on the total (215.7). That much and no more, because over seven seasons the closing line has been the more accurate number.
Ratings only, ignoring the line entirely: BOS -4.9 · 215.7. The model's independent opinion — and the least accurate of the three.
Where these numbers come from
Our model starts from the line and adjusts it for what team ratings, injuries, rest and travel say the line missed. How far it moves is what a pick is based on.
The percentages below use a blend: the line moved 40% of the way toward our model on the spread (IND -3.6) and 5% on the total (233.8). That much and no more, because over seven seasons the closing line has been the more accurate number.
Ratings only, ignoring the line entirely: IND -3.6 · 233.8. The model's independent opinion — and the least accurate of the three.
Where these numbers come from
Our model starts from the line and adjusts it for what team ratings, injuries, rest and travel say the line missed. How far it moves is what a pick is based on.
The percentages below use a blend: the line moved 40% of the way toward our model on the spread (GSW -2.0) and 5% on the total (226.4). That much and no more, because over seven seasons the closing line has been the more accurate number.
Ratings only, ignoring the line entirely: GSW -2.0 · 226.4. The model's independent opinion — and the least accurate of the three.
Where these numbers come from
Our model starts from the line and adjusts it for what team ratings, injuries, rest and travel say the line missed. How far it moves is what a pick is based on.
The percentages below use a blend: the line moved 40% of the way toward our model on the spread (LAL -7.2) and 5% on the total (224.2). That much and no more, because over seven seasons the closing line has been the more accurate number.
Ratings only, ignoring the line entirely: LAL -7.2 · 224.2. The model's independent opinion — and the least accurate of the three.
How to read a card. Line is the sportsbook's number. Our model is the same game after adjusting for team ratings, injuries, rest and travel. A spread or total becomes a play only when our model is at least 3.0 points off the spread or 5.0 off the total — smaller gaps have never beaten the book in seven seasons of testing — and a moneyline only at 10.0% expected value. Our chance is how often we expect the pick to win, next to the book's chance from its own price (with the vig taken out); EV is the expected return per dollar at the posted price. A pick that clears the bar waits (waiting on news) until both teams have played enough games to rate and no unresolved injury could still move it 1.5 points. The number after each injured player is roughly how many points a game he's worth. Prices are from ESPN's partner book.
Record by disagreement
Spreads
| |model − line| | Record | Win% | Units |
|---|---|---|---|
| 0–0.5 pts | 11-15 | 42.3% | +0.00 |
| 0.5–1 pts | 8-11 | 42.1% | +0.00 |
| 1–2 pts | 11-8 | 57.9% | +0.00 |
| 2–3 pts | 7-4 | 63.6% | +0.00 |
| 3+ pts | 3-3 | 50.0% | -0.91 |
Totals
| |model − line| | Record | Win% | Units |
|---|---|---|---|
| 0–1 pts | 15-15 | 50.0% | +0.00 |
| 1–2 pts | 15-12 | 55.6% | +0.00 |
| 2–3 pts | 12-8 | 60.0% | +0.00 |
| 3–4.5 pts | 4-0 | 100.0% | +0.00 |
Moneyline
| EV at the price | Record | Win% | Units |
|---|---|---|---|
| negative | 25-20 | 55.6% vs 65.3 | +0.00 |
| 0–2% | 12-4 | 75.0% vs 64.0 | +0.00 |
| 2–5% | 5-5 | 50.0% vs 48.3 | +0.00 |
| 5–10% | 4-6 | 40.0% vs 53.6 | +0.00 |
Every graded game this season, bucketed by how far the model sat from the line (or, for the moneyline, by EV at the price) — each pick above shows the row it falls in. Green beats the bar and red falls short; a bucket stays uncoloured until it has 20 games.
Recent results
| Game | Spread | Total | Moneyline | Units |
|---|---|---|---|---|
| DEN at PHX 2026-10-30 · 127-120 |
DEN -3.5 win |
Under 231.5 loss |
DEN win |
+0.00 |
| ORL at TOR 2026-10-30 · 121-114 |
TOR -5.5 loss |
Under 227.5 loss |
TOR loss |
+0.00 |
| CHA at WAS 2026-10-30 · 79-95 |
CHA -1.5 loss |
Under 218.5 win |
CHA loss |
-0.91 |
| DET at BKN 2026-10-30 · 110-94 |
DET -7.5 win |
Over 222.5 loss |
DET win |
+0.00 |
| NYK at PHI 2026-10-30 · 93-98 |
NYK -2.5 loss |
Under 227.5 win |
NYK loss |
+0.00 |
| CHI at BOS 2026-10-30 · 111-125 |
BOS -15.5 loss |
Under 236.5 win |
BOS win |
+0.00 |
| IND at MIA 2026-10-30 · 125-145 |
IND +8.5 loss |
Under 237.5 loss |
MIA win |
+0.00 |
| HOU at DAL 2026-10-30 · 111-116 |
HOU -6.5 loss |
Under 223.5 loss |
HOU loss |
+0.00 |
| LAC at MIN 2026-10-30 · 101-127 |
MIN -7.5 win |
Under 223.5 loss |
MIN win |
+0.00 |
| LAL at GSW 2026-10-30 · 95-112 |
LAL -2.5 loss |
Under 234.5 win |
LAL loss |
+0.00 |
| MEM at POR 2026-10-29 · 124-103 |
MEM +7.5 win |
Under 232.5 win |
POR loss |
+0.00 |
| SAC at NOP 2026-10-29 · 120-132 |
NOP -6.5 win |
Over 235.5 win |
NOP win |
+0.00 |