2025 NFL Week 3 Model Backtest: Pick vs Actual
Quick answer: Using only 2024 regular-season
results (no 2025 look-ahead), this model went 10/16 (62.5%) on 2025 Week 3
decided games. Mean Brier 0.1967. Full 2025 regular season:
171/271 (63.1%). These are graded historical projections, not live odds.
This is a backtest recap, not a 2026 preview. Ratings were
fit on completed 2024 REG games only, then frozen. Every row below has a
real final score from nflverse. Ties are excluded from the hit rate.
Highest-confidence hit: Buffalo Bills (84.0%) in Miami Dolphins at Buffalo Bills. Costliest miss: picked Green Bay Packers at 83.5% and lost Green Bay Packers 10–Cleveland Browns 13.
Week 3 scorecard
| Metric | Value |
|---|---|
| Games | 16 |
| Decided (no ties) | 16 |
| Correct picks | 10/16 (62.5%) |
| Mean Brier (home WP vs outcome) | 0.1967 |
| Full 2025 REG accuracy (same frozen model) | 171/271 (63.1%) |
Every Week 3 game: pick vs actual
| Date | Away (score) | Home (score) | Home WP | Pick | Actual | Result | Source |
|---|---|---|---|---|---|---|---|
| 2025-09-18 | Miami Dolphins 21 | Buffalo Bills 31 | 84.0% | Buffalo Bills | Buffalo Bills | HIT | ensemble |
| 2025-09-21 | Atlanta Falcons 0 | Carolina Panthers 30 | 33.3% | Atlanta Falcons | Carolina Panthers | MISS | ensemble |
| 2025-09-21 | Dallas Cowboys 14 | Chicago Bears 31 | 44.6% | Dallas Cowboys | Chicago Bears | MISS | ensemble |
| 2025-09-21 | Green Bay Packers 10 | Cleveland Browns 13 | 16.5% | Green Bay Packers | Cleveland Browns | MISS | ensemble |
| 2025-09-21 | Houston Texans 10 | Jacksonville Jaguars 17 | 39.4% | Houston Texans | Jacksonville Jaguars | MISS | ensemble |
| 2025-09-21 | Denver Broncos 20 | Los Angeles Chargers 23 | 58.7% | Los Angeles Chargers | Los Angeles Chargers | HIT | ensemble |
| 2025-09-21 | Cincinnati Bengals 10 | Minnesota Vikings 48 | 68.8% | Minnesota Vikings | Minnesota Vikings | HIT | ensemble |
| 2025-09-21 | Pittsburgh Steelers 21 | New England Patriots 14 | 36.0% | Pittsburgh Steelers | Pittsburgh Steelers | HIT | ensemble |
| 2025-09-21 | Kansas City Chiefs 22 | New York Giants 9 | 17.1% | Kansas City Chiefs | Kansas City Chiefs | HIT | ensemble |
| 2025-09-21 | Los Angeles Rams 26 | Philadelphia Eagles 33 | 70.4% | Philadelphia Eagles | Philadelphia Eagles | HIT | ensemble |
| 2025-09-21 | New Orleans Saints 13 | Seattle Seahawks 44 | 77.5% | Seattle Seahawks | Seattle Seahawks | HIT | ensemble |
| 2025-09-21 | Arizona Cardinals 15 | San Francisco 49ers 16 | 48.0% | Arizona Cardinals | San Francisco 49ers | MISS | ensemble |
| 2025-09-21 | New York Jets 27 | Tampa Bay Buccaneers 29 | 77.2% | Tampa Bay Buccaneers | Tampa Bay Buccaneers | HIT | ensemble |
| 2025-09-21 | Indianapolis Colts 41 | Tennessee Titans 20 | 27.4% | Indianapolis Colts | Indianapolis Colts | HIT | ensemble |
| 2025-09-21 | Las Vegas Raiders 24 | Washington Commanders 41 | 71.4% | Washington Commanders | Washington Commanders | HIT | ensemble |
| 2025-09-22 | Detroit Lions 38 | Baltimore Ravens 30 | 51.2% | Baltimore Ravens | Detroit Lions | MISS | ensemble |
Misses this week
Sorted by pick confidence. A toss-up miss is cheaper on Brier than a
heavy favorite that lost.
| Game | Pick | Pick WP | Actual | Margin |
|---|---|---|---|---|
| Green Bay Packers 10 at Cleveland Browns 13 | Green Bay Packers | 83.5% | Cleveland Browns | 3 |
| Atlanta Falcons 0 at Carolina Panthers 30 | Atlanta Falcons | 66.7% | Carolina Panthers | 30 |
| Houston Texans 10 at Jacksonville Jaguars 17 | Houston Texans | 60.6% | Jacksonville Jaguars | 7 |
| Dallas Cowboys 14 at Chicago Bears 31 | Dallas Cowboys | 55.4% | Chicago Bears | 17 |
| Arizona Cardinals 15 at San Francisco 49ers 16 | Arizona Cardinals | 52.0% | San Francisco 49ers | 1 |
| Detroit Lions 38 at Baltimore Ravens 30 | Baltimore Ravens | 51.2% | Detroit Lions | 8 |
How to read HIT / MISS
Pick is the side with win probability ≥ 50%. A 51–49 toss-up that loses
is a MISS the same as an 80–20 favorite that loses — the Brier score is
the one that punishes overconfidence. Ensemble means a vig-free 2025
moneyline was blended 60/40 with the 2024 ratings; model means ratings
only. Lines here are the historical nflverse close, not a current book.
The frozen 2024 prior went 171/271 (63.1%) across the whole 2025 regular season
(271 decided games, one tie). Week 3 is one slice of that grade.
We start here because early weeks are the honest test of a prior-season
rating: no in-season update, no injury news, no “we always knew.”
Related:
2026 game predictions
(same formula, 2025 ratings) and the
NFL Stats Hub.
Weeks 1–4 at a glance (same frozen 2024 model)
| Week | Correct | Decided | Hit rate | Mean Brier |
|---|---|---|---|---|
| 1 | 12 | 16 | 75.0% | 0.1879 |
| 2 | 10 | 16 | 62.5% | 0.1879 |
| 3 (this page) | 10 | 16 | 62.5% | 0.1967 |
| 4 | 11 | 15 | 73.3% | 0.2042 |
Week 1 was the high-water mark in this opening stretch. Accuracy dipped
in Weeks 2–3, which is what a frozen prior-season rating should do once
the new year starts producing its own information. We did not peek at
those later results to retune anything on this page.
Methodology (no look-ahead)
- Train: every completed 2024 regular-season game. Playoffs excluded.
- Test: 2025 Week 3 completed regular-season games.
- Same rating formula as the live 2026 sheet: logit(win%) + 0.01×(PPG−PPG allowed), plus 2024 empirical home-field.
- Fail-closed: if 2024 is incomplete or a 2025 game has no score, this page is not built from mocks.
FAQ
Did this model exist in September 2025?
No. This is a retrospective run of the same formula we use for 2026,
frozen on 2024 data. It is an honest grade, not a claim we published
these picks live last year.
Why start with Weeks 1–4?
Early weeks are the hardest test of a prior-season rating. Later weeks
would look better if we updated ratings mid-season — we did not.
Is this betting advice?
No. Historical model grade for nflfacts.com stats pages.
What does Brier 0.19 actually mean?
Brier is the squared error of the home win probability versus the 0/1
outcome. Coin-flip guesses land near 0.25. A 0.19 week is better than
guessing and worse than a sharp closer. We publish it so a 12/16 week
cannot hide four confident misses.
If you only remember one number from this page, use the hit rate for
the story and the Brier for the honesty check. Both are computed from
the same frozen 2024 ratings. Neither uses 2025 results as an input.
Sources
Backtest rebuilt 2026-09-04 UTC. Train 2024 REG, test 2025 Week 3.
Not a live box score.