Model Accuracy
Every Game Intel page leads with ChallengeFlag’s win probability, published before kickoff and never changed after it. This page keeps score. The pick is the side ChallengeFlag gives better than even odds. A tied game, or one it calls at exactly 50%, counts as half a win and shows as the third number in the record.
The Brier score grades the probabilities, not just the picks: lower is better, and calling every game 50% scores 0.250. A pick at 80% that loses costs more than a pick at 55% that loses.
Pro, by week
| Week | Record | Right |
|---|---|---|
| Week 1 | 8-8 | 50.0% |
| Week 2 | 9-7 | 56.2% |
| Week 3 | 9-7 | 56.2% |
| Week 4 | 13-3 | 81.2% |
| Week 5 | 8-6-1 | 56.7% |
| Week 6 | 0-1 | 0.0% |
College, by week
| Week | Record | Right |
|---|---|---|
| Week 6 | 38-20 | 65.5% |
| Week 7 | 4-3 | 57.1% |
How the forecasts are made
Each forecast weighs team strength, recent form and game conditions into a single win probability for each side. It is published before kickoff and never changed after.
These are model outputs, not advice. A season of record is a small sample, and a good week can flatter a model as easily as a bad one can hurt it. ChallengeFlag does not take or place bets.
The draft model
Draft rankings are off between drafts and return for the 2027 class. The 2026 review stays here.
Draft model: the 2026 class, before and after the draft
Every prediction below was published before the 2026 draft. Draft position stands in for the NFL’s evaluation until careers mature; career-outcome validation for the class comes in 2029 and 2030. The model looks for patterns in historical data that have separated stars from busts, then applies them to the new class.
| Position | Player | p(Star) | Confidence | Actual pick | Result |
|---|---|---|---|---|---|
| QB | Fernando Mendoza, Indiana | 54.5% | Low | #1 Las Vegas | Hit |
| WR | Carnell Tate, Ohio State | 38.5% | Medium | #4 Tennessee | Hit |
| LB | Arvell Reese, Ohio State | 49.6% | High | #5 New York Giants | Hit |
| OT | Spencer Fano, Utah | 49.6% | High | #9 Cleveland | Hit |
| DE | Malachi Lawrence, UCF | 51.0% | High | #23 Dallas | Late first round |
| CB | Treydan Stukes, Arizona | 48.9% | High | #38 Las Vegas | Second round |
| S | Lorenzo Styles Jr., Ohio State | 49.8% | High | #172 New Orleans | Miss |
The 2022 to 2024 backtest found the model much better at spotting busts (81 to 90% accuracy on non-skill “Dud” calls) than stars (0% star recall), a known limitation going into 2026 and the focus of the 2027 model.