2025 College Football Week 4 Model Backtest
Quick answer: In this retrospective 2025
Week 4 FBS-vs-FBS backtest, the frozen model went 36/50 (72.0%) with mean Brier
0.1939. Full 2025: 550/762 (72.2%). These were not live picks.
Ratings use final 2023-2024 FBS-vs-FBS games only. They were frozen before any
2025 result was graded. Historical pregame spreads may be blended, but their probability
mapping was calibrated on 2024 outcomes—not on the 2025 test results.
Week 4 scorecard
| Metric | Value |
|---|---|
| Games | 50 |
| Decided | 50 |
| Correct picks | 36/50 (72.0%) |
| Mean Brier | 0.1939 |
| Full-season accuracy | 550/762 (72.2%) |
Full-season method comparison
The headline is the blended ensemble. This table separates the ratings model from
the historical spread so market information is not misreported as pure model performance.
| Method | Games | Correct | Accuracy | Mean Brier |
|---|---|---|---|---|
| Elo model only | 762 | 495 | 65.0% | 0.221 |
| Market spread only | 762 | 564 | 74.0% | 0.1733 |
| 40/60 ensemble | 762 | 550 | 72.2% | 0.1804 |
Every Week 4 FBS-vs-FBS game: pick vs actual
| Date | Away | Home | Home WP | Pick | Actual | Result | Source |
|---|---|---|---|---|---|---|---|
| 2025-09-18 | Rice 28 | Charlotte 17 | 47.7% | Rice | Rice | HIT | ensemble |
| 2025-09-19 | Tulsa 19 | Oklahoma State 12 | 78.5% | Oklahoma State | Tulsa | MISS | ensemble |
| 2025-09-20 | Iowa 38 | Rutgers 28 | 38.5% | Iowa | Iowa | HIT | ensemble |
| 2025-09-20 | North Texas 45 | Army 38 | 59.2% | Army | North Texas | MISS | ensemble |
| 2025-09-20 | Syracuse 34 | Clemson 21 | 82.2% | Clemson | Syracuse | MISS | ensemble |
| 2025-09-20 | Bowling Green 17 | Louisville 40 | 86.2% | Louisville | Louisville | HIT | ensemble |
| 2025-09-20 | Arkansas 31 | Memphis 32 | 52.2% | Memphis | Memphis | HIT | ensemble |
| 2025-09-20 | UNLV 41 | Miami (OH) 38 | 50.1% | Miami (OH) | UNLV | MISS | ensemble |
| 2025-09-20 | SMU 24 | TCU 35 | 51.4% | TCU | TCU | HIT | ensemble |
| 2025-09-20 | Texas Tech 34 | Utah 10 | 55.4% | Utah | Texas Tech | MISS | ensemble |
| 2025-09-20 | Maryland 27 | Wisconsin 10 | 73.9% | Wisconsin | Maryland | MISS | ensemble |
| 2025-09-20 | UAB 24 | Tennessee 56 | 96.6% | Tennessee | Tennessee | HIT | ensemble |
| 2025-09-20 | Oregon State 7 | Oregon 41 | 97.0% | Oregon | Oregon | HIT | ensemble |
| 2025-09-20 | Troy 21 | Buffalo 17 | 59.9% | Buffalo | Troy | MISS | ensemble |
| 2025-09-20 | Louisiana 31 | Eastern Michigan 34 | 37.0% | Louisiana | Eastern Michigan | MISS | ensemble |
| 2025-09-20 | Kent State 10 | Florida State 66 | 96.2% | Florida State | Florida State | HIT | ensemble |
| 2025-09-20 | James Madison 31 | Liberty 13 | 46.3% | James Madison | James Madison | HIT | ensemble |
| 2025-09-20 | Michigan 30 | Nebraska 27 | 38.0% | Michigan | Michigan | HIT | ensemble |
| 2025-09-20 | Purdue 30 | Notre Dame 56 | 95.0% | Notre Dame | Notre Dame | HIT | ensemble |
| 2025-09-20 | Auburn 17 | Oklahoma 24 | 69.7% | Oklahoma | Oklahoma | HIT | ensemble |
| 2025-09-20 | Tulane 10 | Ole Miss 45 | 75.1% | Ole Miss | Ole Miss | HIT | ensemble |
| 2025-09-20 | North Carolina 9 | UCF 34 | 58.2% | UCF | UCF | HIT | ensemble |
| 2025-09-20 | Ball State 25 | UConn 31 | 86.6% | UConn | UConn | HIT | ensemble |
| 2025-09-20 | Toledo 13 | Western Michigan 14 | 30.8% | Toledo | Western Michigan | MISS | ensemble |
| 2025-09-20 | NC State 33 | Duke 45 | 64.3% | Duke | Duke | HIT | ensemble |
| 2025-09-20 | Northern Illinois 10 | Mississippi State 38 | 65.6% | Mississippi State | Mississippi State | HIT | ensemble |
| 2025-09-20 | Temple 24 | Georgia Tech 45 | 92.2% | Georgia Tech | Georgia Tech | HIT | ensemble |
| 2025-09-20 | Delaware 38 | Florida International 16 | 61.2% | Florida International | Delaware | MISS | ensemble |
| 2025-09-20 | West Virginia 10 | Kansas 41 | 76.3% | Kansas | Kansas | HIT | ensemble |
| 2025-09-20 | Arkansas State 21 | Kennesaw State 28 | 36.1% | Arkansas State | Kennesaw State | MISS | ensemble |
| 2025-09-20 | Boise State 49 | Air Force 37 | 19.3% | Boise State | Boise State | HIT | ensemble |
| 2025-09-20 | Marshall 42 | Middle Tennessee 28 | 30.7% | Marshall | Marshall | HIT | ensemble |
| 2025-09-20 | South Carolina 20 | Missouri 29 | 67.2% | Missouri | Missouri | HIT | ensemble |
| 2025-09-20 | Coastal Carolina 38 | South Alabama 20 | 76.2% | South Alabama | Coastal Carolina | MISS | ensemble |
| 2025-09-20 | Nevada 16 | Western Kentucky 31 | 77.8% | Western Kentucky | Western Kentucky | HIT | ensemble |
| 2025-09-20 | Arizona State 27 | Baylor 24 | 48.8% | Arizona State | Arizona State | HIT | ensemble |
| 2025-09-20 | BYU 34 | East Carolina 13 | 34.1% | BYU | BYU | HIT | ensemble |
| 2025-09-20 | Illinois 10 | Indiana 63 | 70.2% | Indiana | Indiana | HIT | ensemble |
| 2025-09-20 | Southern Miss 20 | Louisiana Tech 30 | 68.1% | Louisiana Tech | Louisiana Tech | HIT | ensemble |
| 2025-09-20 | Florida 7 | Miami 26 | 71.3% | Miami | Miami | HIT | ensemble |
| 2025-09-20 | Georgia State 21 | Vanderbilt 70 | 86.2% | Vanderbilt | Vanderbilt | HIT | ensemble |
| 2025-09-20 | Stanford 20 | Virginia 48 | 78.7% | Virginia | Virginia | HIT | ensemble |
| 2025-09-20 | Washington 59 | Washington State 24 | 21.8% | Washington | Washington | HIT | ensemble |
| 2025-09-21 | Sam Houston 0 | Texas 55 | 95.5% | Texas | Texas | HIT | ensemble |
| 2025-09-21 | UL Monroe 31 | UTEP 25 | 63.3% | UTEP | UL Monroe | MISS | ensemble |
| 2025-09-21 | UTSA 17 | Colorado State 16 | 41.9% | UTSA | UTSA | HIT | ensemble |
| 2025-09-21 | Wyoming 20 | Colorado 37 | 81.1% | Colorado | Colorado | HIT | ensemble |
| 2025-09-21 | California 0 | San Diego State 34 | 19.9% | California | San Diego State | MISS | ensemble |
| 2025-09-21 | Michigan State 31 | USC 45 | 86.0% | USC | USC | HIT | ensemble |
| 2025-09-21 | Fresno State 23 | Hawai’i 21 | 40.5% | Fresno State | Fresno State | HIT | ensemble |
Highest-confidence misses
Limited to the 25 most confident misses so the table
stays readable; the complete game-level record is preserved in the JSON artifact.
| Game | Pick | Confidence | Actual | Margin |
|---|---|---|---|---|
| Syracuse at Clemson | Clemson | 82.2% | Syracuse | 13 |
| California at San Diego State | California | 80.2% | San Diego State | 34 |
| Tulsa at Oklahoma State | Oklahoma State | 78.5% | Tulsa | 7 |
| Coastal Carolina at South Alabama | South Alabama | 76.2% | Coastal Carolina | 18 |
| Maryland at Wisconsin | Wisconsin | 73.9% | Maryland | 17 |
| Toledo at Western Michigan | Toledo | 69.2% | Western Michigan | 1 |
| Arkansas State at Kennesaw State | Arkansas State | 63.9% | Kennesaw State | 7 |
| UL Monroe at UTEP | UTEP | 63.3% | UL Monroe | 6 |
| Louisiana at Eastern Michigan | Louisiana | 63.0% | Eastern Michigan | 3 |
| Delaware at Florida International | Florida International | 61.2% | Delaware | 22 |
| Troy at Buffalo | Buffalo | 59.9% | Troy | 4 |
| North Texas at Army | Army | 59.2% | North Texas | 7 |
| Texas Tech at Utah | Utah | 55.4% | Texas Tech | 24 |
| UNLV at Miami (OH) | Miami (OH) | 50.1% | UNLV | 3 |
Methodology and interpretation
- Train: 2023 and 2024 final regular-season FBS-vs-FBS games.
- Test: 2025 final regular-season FBS-vs-FBS games, with ratings frozen.
- Model: opponent-adjusted Elo with score-margin adjustment and empirical home field.
- Ensemble rows: 40% Elo plus 60% historical spread-derived win probability.
- Brier measures probability error; lower is better and 0.25 is the coin-flip reference.
FAQ
Did this model exist in the 2025 season?
No. This is a retrospective reconstruction, clearly labeled as such. It does not claim
these picks were published before kickoff.
Why only FBS-vs-FBS?
It gives one comparable
opponent pool and avoids pretending schedules across NCAA divisions are interchangeable.
Is this betting advice?
No. It is a model audit and historical scorecard.
Related: 2026 CFB predictions
and the NFL Facts Stats Hub.
Sources
Retrospective backtest rebuilt 2026-09-06.
No mock data and no 2025 result used to fit the frozen ratings.