Our 2026 projections are a pure model, zero betting-market input. This is the honest side-by-side: how closely the model tracks the market across every season surface, and the teams where it most diverges. Where we differ is where a model earns its keep, or where the market knows something our numbers can't.
Correlation with the betting market, 0–1 (higher = more aligned). Our pure model tracks the market at 0.80 on wins and picks the same game favorite 83% of the time, credible, but with its own opinions. The model’s win-total spread is 71% of the market’s: a pure preseason model is deliberately more conservative about calling a team great or terrible.
Green = our model is higher than the market; red = lower. These are the teams to watch: either the model sees value the market hasn’t priced, or the market knows something an EPA-based model can’t (a QB situation, a gutted roster). We show it rather than quietly bend our number toward the line.
| TEAM | WINS (M) | WINS (MKT) | PLAYOFF (M) | SB (M) |
|---|---|---|---|---|
| 10.3 | 10.2 | 78% | 10.9% | |
| 10.2 | 10.3 | 69% | 7.3% | |
| 10.1 | 9.9 | 70% | 6.1% | |
| 9.9 | 10.8 | 67% | 6.7% | |
| 9.9 | 10.6 | 68% | 6.3% | |
| 9.8 | 9.5 | 68% | 5.1% | |
| 9.5 | 9.8 | 58% | 5.2% | |
| 9.3 | 9.8 | 60% | 4.8% | |
| 9.3 | 10.6 | 55% | 3.9% | |
| 9.1 | 9.4 | 57% | 3.7% | |
| 9.1 | 9.9 | 55% | 4.0% | |
| 9.1 | 9.8 | 50% | 3.6% | |
| 9.0 | 8.9 | 54% | 3.4% | |
| 9.0 | 7.9 | 53% | 3.1% | |
| 8.8 | 8.3 | 49% | 2.6% | |
| 8.8 | 8.5 | 44% | 2.5% | |
| 8.6 | 9.5 | 46% | 2.9% | |
| 8.4 | 8.2 | 42% | 2.2% | |
| 8.4 | 9.3 | 39% | 2.9% | |
| 8.4 | 6.8 | 41% | 2.3% | |
| 8.4 | 7.8 | 38% | 2.0% | |
| 8.2 | 7.7 | 35% | 1.3% | |
| 8.1 | 6.5 | 36% | 1.4% | |
| 7.9 | 10.0 | 31% | 1.1% | |
| 7.8 | 9.0 | 28% | 1.2% | |
| 7.5 | 4.5 | 22% | 0.9% | |
| 7.2 | 7.6 | 20% | 0.7% | |
| 7.1 | 5.0 | 23% | 1.0% | |
| 6.6 | 7.1 | 15% | 0.4% | |
| 6.1 | 6.2 | 10% | 0.2% | |
| 6.1 | 6.0 | 10% | 0.2% | |
| 6.0 | 6.7 | 10% | 0.2% |
(M) = our model · (MKT/V) = the market. Market numbers are derived by fitting team ratings to the 272 DK opening lines and running the identical simulation, so it’s the same engine, our ratings vs the market’s. Our model never sees a betting line. The WINS (MKT) column here is derived from individual game lines, not the same number as the real posted DraftKings season win total, and the two can differ by a few tenths of a win.
Every Wednesday the model picks every game and emails the slate before kickoff. Every pick is graded in public, wins and losses.