Scorecard
Our last number before kickoff, graded against the final score and the closing line.
2026 season only
Frozen pregame record
A scheduled job saves the model's current number for every upcoming game (at most once an hour, with the market line at that moment) and stops at kickoff. Each game below is graded from the last number saved before it kicked off, so the result could never have changed it. Closing line = the last market consensus this site stored before kickoff, else the nflverse closing line. 16 upcoming games have a snapshot; newest Oct 3, 3:29 AM ET.
Closing line, same games: 8/16 winners · 7.1 points off the margin · 11.5 off the total (16 games with a recorded closing line).
Our margin was closer to the final than the closing line in 63% of games; on average our number sat 2.1 points from the closing line. Brier score 0.209 (0.250 = coin flip).
Our total vs what they scored
When we said the total would be Under 40 points
2 games · we said 39.7 on average · they really scored 35.0
When we said the total would be 40 to 43.9 points
5 games · we said 42.4 on average · they really scored 48.0
When we said the total would be 44 to 47.9 points
6 games · we said 45.9 on average · they really scored 48.7
When we said the total would be 48 to 51.9 points
2 games · we said 51.3 on average · they really scored 45.5
When we said the total would be 52 to 55.9 points
1 game · we said 53.0 on average · they really scored 65.0
Our winning margin vs the real one
When we projected A win by less than 3
8 games · we said the winner would win by 1.0 · the real margin was 9.0
Right winner in 4 out of 8 of them (over a long run, expect between 22% and 78%).
When we projected A win by 3 to 6.9
4 games · we said the winner would win by 5.0 · the real margin was 7.3
Right winner in 4 out of 4 of them (over a long run, expect between 51% and 100%).
When we projected A win by 7 to 10.9
3 games · we said the winner would win by 8.6 · the real margin was 5.7
Right winner in 2 out of 3 of them (over a long run, expect between 21% and 94%).
When we projected A win by 11 to 13.9
1 game · we said the winner would win by 11.8 · the real margin was 6.0
Right winner in 1 out of 1 of them (over a long run, expect between 21% and 100%).
Every graded game
Week 4 · we said CLE by 0.7, total 39.8 · close PIT by 2.5, 38
Frozen Oct 1, 7:52 PM ET
Week 3 · we said PHI by 1.6, total 44.5 · close PHI by 3.5, 42.5
Frozen Sep 28, 7:52 PM ET
Week 3 · we said LA by 0.5, total 44.4 · close LA by 1.5, 44
Frozen Sep 27, 7:52 PM ET
Week 3 · we said BAL by 4.0, total 53.0 · close BAL by 3.5, 53
Frozen Sep 27, 3:52 PM ET
Week 3 · we said NO by 2.9, total 45.7 · close NO by 3.5, 43.5
Frozen Sep 27, 3:52 PM ET
Week 3 · we said SF by 11.8, total 47.7 · close SF by 7.5, 48
Frozen Sep 27, 3:52 PM ET
Week 3 · we said MIN by 4.9, total 42.8 · close MIN by 1.5, 42.5
Frozen Sep 27, 3:52 PM ET
Week 3 · we said BUF by 8.7, total 47.5 · close BUF by 7.0, 50
Frozen Sep 27, 12:52 PM ET
Week 3 · we said CLE by 0.7, total 41.7 · close CAR by 2.5, 42
Frozen Sep 27, 12:52 PM ET
Week 3 · we said DET by 8.0, total 51.5 · close DET by 6.5, 48.5
Frozen Sep 27, 12:52 PM ET
Week 3 · we said IND by 0.1, total 51.1 · close HOU by 1.5, 42.5
Frozen Sep 27, 12:52 PM ET
Week 3 · we said KC by 6.5, total 45.6 · close KC by 10.0, 45.5
Frozen Sep 27, 12:52 PM ET
Week 3 · we said NYG by 4.4, total 39.6 · close NYG by 2.5, 37.5
Frozen Sep 27, 12:52 PM ET
Week 3 · we said CIN by 0.5, total 42.8 · close CIN by 3.0, 42.5
Frozen Sep 27, 12:52 PM ET
Week 3 · we said SEA by 9.0, total 42.3 · close SEA by 8.5, 40.5
Frozen Sep 27, 12:52 PM ET
Week 3 · we said JAX by 0.9, total 42.5 · close JAX by 3.0, 46
Frozen Sep 27, 12:52 PM ET
Legacy numbers (not in the headline)
Before the hourly snapshots, the site kept the first number any visitor happened to see, from an older version of the model, sometimes days before kickoff. They are shown here for the record but left out of every headline number above. 33% winners (5/15), 15.7 points off the margin, 12.6 off the total; closing line on the same games 13.2 off the margin.
Week 3 · we said GB by 2.9, total 44.7 · close GB by 4.5, 43.5
First seen Sep 20, 12:57 PM ET
Week 2 · we said NYG by 2.3, total 43.1 · close LA by 6.5, 47.5
First seen Sep 19, 10:13 AM ET
Week 2 · we said KC by 9.5, total 49.1 · close KC by 6.0, 46.5
First seen Sep 19, 10:13 AM ET
Week 2 · we said DAL by 1.0, total 45.0 · close DAL by 3.5, 51.5
First seen Sep 19, 10:13 AM ET
Week 2 · we said ARI by 3.4, total 36.2 · close SEA by 3.5, 40.5
First seen Sep 19, 10:13 AM ET
Week 2 · we said SF by 6.9, total 39.5 · close SF by 12.5, 44.5
First seen Sep 19, 10:13 AM ET
Week 2 · we said JAX by 6.4, total 40.2 · close DEN by 2.5, 45.5
First seen Sep 19, 10:13 AM ET
Week 2 · we said LV by 3.3, total 39.7 · close LAC by 6.5, 43.5
First seen Sep 19, 10:13 AM ET
Week 2 · we said ATL by 2.4, total 55.0 · close CAR by 2.5, 43.5
First seen Sep 19, 10:13 AM ET
Week 2 · we said CHI by 5.8, total 65.4 · close CHI by 4.5, 46.5
First seen Sep 19, 10:13 AM ET
Week 2 · we said PHI by 1.9, total 40.5 · close PHI by 7.0, 39.5
First seen Sep 19, 10:13 AM ET
Week 2 · we said PIT by 0.0, total 34.4 · close NE by 4.5, 41.5
First seen Sep 19, 10:13 AM ET
Week 2 · we said NYJ by 6.2, total 47.4 · close GB by 3.5, 44.5
First seen Sep 19, 10:13 AM ET
Week 2 · we said TB by 5.1, total 44.3 · close TB by 8.5, 41.5
First seen Sep 19, 10:13 AM ET
Week 2 · we said BAL by 10.2, total 58.0 · close BAL by 8.5, 45.5
First seen Sep 19, 10:13 AM ET
Week 2 · we said HOU by 2.1, total 55.1 · close HOU by 3.0, 45.5
First seen Sep 19, 10:13 AM ET
Reference table from earlier tracked games (not part of the frozen record above):
Total scoring guide
What the model total has meant in tracked games.
| We said | Call | Record |
|---|---|---|
| 38 or less | Under 41Strongest | 8–0 (100%) |
| 44 or less | Under 45 | 14–1 (93%) |
| 45 to 54 | Over 45 | 6–1 (86%) |
| 55+ | Skip | 5–4 |
The first two rows overlap. Every game at 38 or less is also in the 44-or-less row. When both apply, Under 41 is the stronger play.
2026 regular-season finals only. The walk-forward backtest, which recomputes every game with the current model, is on the methodology page and is labelled as a backtest there.