The angle: The model makes it Ole Miss by 8.9; DraftKings has Missouri +27.5, a 18.6-point disagreement toward Missouri. Driven by: Missouri's defense ranks 6th in FBS in success rate allowed to Ole Miss's 81st (worth 6.9 pts); Missouri's offense ranks 42nd in FBS in explosiveness to Ole Miss's 115th (worth 1.2 pts). Injury report: Ole Miss lists 9 players out or doubtful (LB Ant Davis Jr., OL Emanuel Faulkner, OL Enoch Wangoy, ...). Risk: Missouri lists 6 players out or doubtful (OL Josh Atkins, OL Khalief Canty Jr., RB Ahmad Hardy, ...).
College Football Model
A point-in-time CFB forecasting system that publishes its own spread and total against verified DraftKings quotes, with opponent-adjusted team strength, early-season reliability weighting, and a timestamped shadow record.
Week 7 Best Bets
The board’s strongest disagreements with DraftKings, each with the angle behind it. Each pick is graded at the line it was first published at. Hit estimates give the model only the weight it earned against the market in held-out seasons, so they sit close to 50%. Every pick is logged before kickoff and graded. Season record: Spreads 10-8-0 (+1.1u, mean CLV +0.56) · Totals 4-5-0 (-1.4u, mean CLV +1.22). Evidence bar: 27 of 200 graded picks, win rate 51.9% (95% lower bound 34.0% vs 52.4% breakeven) - not met; these stay research picks. Research picks, not advice: authority is RESEARCH_ONLY until the record earns otherwise.
Spreads
The angle: The model makes it Toledo by 3.2; DraftKings has Eastern Michigan +19.5, a 16.3-point disagreement toward Eastern Michigan. Driven by: Eastern Michigan's defense ranks 32nd in FBS in explosiveness allowed to Toledo's 115th (worth 2.0 pts); Eastern Michigan's offense ranks 70th in FBS in success rate to Toledo's 91st (worth 1.2 pts). Injury report: Toledo lists 12 players out or doubtful (DE Langston Long, DE Zackery Samuel, DT Cam Gwinn, ...). Risk: Eastern Michigan QB Noah Kim is listed questionable; Eastern Michigan lists 21 players out or doubtful (DB Antonio Solares-Vitti, DB Jabari Monds, DB Jaivian Norman, ...).
The angle: The model makes it Wyoming by 7.8; DraftKings has Northern Illinois +18.5, a 10.7-point disagreement toward Northern Illinois. Driven by: Northern Illinois's defense ranks 75th in FBS in run-stuff rate to Wyoming's 132nd (worth 1.1 pts); Northern Illinois's offense ranks 90th in FBS in explosiveness to Wyoming's 123rd (worth 0.7 pts). Home field is worth 7.9 pts at this venue (more than the FBS-average 4.5: altitude, travel and time zones).
The angle: The model makes it New Mexico by 23.9; DraftKings has New Mexico -14.5, a 9.4-point disagreement toward New Mexico. Driven by: New Mexico's defense ranks 48th in FBS in success rate allowed to Hawai'i's 137th (worth 6.9 pts); New Mexico's offense ranks 6th in FBS in explosiveness to Hawai'i's 122nd (worth 2.4 pts). Home field is worth 2.4 pts at this venue (less than the FBS-average 4.5: altitude, travel and time zones).
The angle: The model makes it Louisiana by 7.0; DraftKings has Troy +15.5, a 8.5-point disagreement toward Troy. Driven by: Troy's offense ranks 54th in FBS in success rate to Louisiana's 110th (worth 3.8 pts).
Totals
The angle: The model projects 52.9 combined points against a DraftKings total of 34.5 (18.4 over). Efficiency: Missouri offense 52nd / defense 20th in EPA per play; Ole Miss offense 4th / defense 67th in EPA per play. Pace: the two teams average 61 plays a game.
The angle: The model projects 42.5 combined points against a DraftKings total of 56.5 (14.0 under). Efficiency: Washington State offense 128th / defense 60th in EPA per play; Oregon State offense 23rd / defense 31st in EPA per play. Pace: the two teams average 57 plays a game.
The angle: The model projects 51.4 combined points against a DraftKings total of 64.5 (13.1 under). Efficiency: Northern Illinois offense 130th / defense 127th in EPA per play; Wyoming offense 103rd / defense 130th in EPA per play. Pace: the two teams average 61 plays a game.
Week 7 Board
Forecast margin is the home side and comes from the independent model. DraftKings is the live benchmark it is trying to beat, not the published number. Expand any game for the full breakdown. Authority remains RESEARCH_ONLY: historical ATS on disagreements has not cleared 52.38% breakeven.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Middle Tennessee. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Jacksonville State. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Missouri State. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Sam Houston. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours UAB. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Old Dominion. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Texas State. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Tulane. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Coastal Carolina. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Purdue. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Bowling Green. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours West Virginia. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Texas. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Army. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours James Madison. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Kansas State. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours LSU. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Michigan State. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Indiana. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Boston College. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours SMU. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Maryland. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Temple. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Buffalo. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Central Michigan. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Miami (OH). Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Tennessee. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Texas Tech. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Georgia. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Virginia Tech. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Ole Miss. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Duke. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Wyoming. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Michigan. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Eastern Michigan. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Air Force. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Rice. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Louisiana. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Oregon State. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Vanderbilt. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Southern Miss. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Miami. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours South Florida. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours North Dakota State. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Baylor. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Oklahoma. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Syracuse. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours BYU. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Houston. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours UCLA. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours UL Monroe. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours UTSA. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours UTEP. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Sacramento State. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Colorado. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours San Diego State. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Hawai'i. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours California. Allowed statistics are inverted, so a lower number is a better defence.


Breakdown · how the model got there
Values are opponent-adjusted, so they are comparable across schedules. Pts is each factor’s effect on the home margin: positive favours Oregon. Allowed statistics are inverted, so a lower number is a better defence.
Power Ratings
Points relative to an average FBS team on a neutral field, so the projected neutral margin between two teams is the difference of their ratings.
| # | Team | Conference | Rating | vs average |
|---|---|---|---|---|
| 1 | Alabama | SEC | ||
| 2 | Notre Dame | FBS Independents | ||
| 3 | Miami | ACC | ||
| 4 | Ohio State | Big Ten | ||
| 5 | LSU | SEC | ||
| 6 | Georgia | SEC | ||
| 7 | Indiana | Big Ten | ||
| 8 | Ole Miss | SEC | ||
| 9 | Texas | SEC | ||
| 10 | UCLA | Big Ten | ||
| 11 | Houston | Big 12 | ||
| 12 | Utah | Big 12 | ||
| 13 | Oregon | Big Ten | ||
| 14 | Texas Tech | Big 12 | ||
| 15 | BYU | Big 12 | ||
| 16 | Tennessee | SEC | ||
| 17 | Missouri | SEC | ||
| 18 | Florida | SEC | ||
| 19 | Wake Forest | ACC | ||
| 20 | Nebraska | Big Ten | ||
| 21 | South Carolina | SEC | ||
| 22 | Kentucky | SEC | ||
| 23 | Baylor | Big 12 | ||
| 24 | Northwestern | Big Ten | ||
| 25 | Wisconsin | Big Ten | ||
| 26 | James Madison | Sun Belt | ||
| 27 | New Mexico | Mountain West | ||
| 28 | Oklahoma | SEC | ||
| 29 | Auburn | SEC | ||
| 30 | Louisville | ACC | ||
| 31 | USC | Big Ten | ||
| 32 | Mississippi State | SEC | ||
| 33 | Boise State | Pac-12 | ||
| 34 | Arizona | Big 12 | ||
| 35 | Oklahoma State | Big 12 | ||
| 36 | Iowa | Big Ten | ||
| 37 | Texas A&M | SEC | ||
| 38 | Oregon State | Pac-12 | ||
| 39 | UCF | Big 12 | ||
| 40 | North Carolina | ACC |
Conference Ratings
Depth is median minus mean: positive is a deep league, negative means the mean is carried by its top teams. x-conf counts the completed out-of-conference games informing the rating — cross-conference scheduling is sparse, so a league’s level is less firmly identified than the ordering inside it.
| Conference | Teams | Mean | Median | Top 4 | Depth | x-conf | Best | |
|---|---|---|---|---|---|---|---|---|
| 1 | SEC | 16 | +10.01 | +18.01 | -0.22 | 30 | Alabama (+21.9) | |
| 2 | Big Ten | 18 | +5.52 | +15.42 | -0.42 | 41 | Ohio State (+19.1) | |
| 3 | Big 12 | 16 | +5.18 | +12.91 | -0.17 | 31 | Houston (+14.0) | |
| 4 | ACC | 17 | +3.23 | +10.82 | +0.33 | 30 | Miami (+19.3) | |
| 5 | Pac-12 | 8 | -2.90 | +4.26 | -1.16 | 25 | Boise State (+8.0) | |
| 6 | American Athletic | 14 | -3.49 | +2.01 | +0.07 | 38 | North Texas (+3.7) | |
| 7 | Sun Belt | 14 | -6.04 | +1.52 | -0.40 | 40 | James Madison (+8.5) | |
| 8 | Mountain West | 10 | -9.70 | +2.18 | -2.50 | 26 | New Mexico (+8.3) | |
| 9 | Conference USA | 10 | -8.60 | -3.08 | -0.97 | 30 | Liberty (+2.5) | |
| 10 | Mid-American | 13 | -9.70 | -3.54 | +0.01 | 34 | Western Michigan (+0.1) |
SEC — 16 teams
- 1Alabama+21.89
- 2LSU+18.04
- 3Georgia+17.03
- 4Ole Miss+15.07
- 5Texas+14.29
- 6Tennessee+10.63
- 7Missouri+10.47
- 8Florida+10.27
- 9South Carolina+9.75
- 10Kentucky+9.47
- 11Oklahoma+8.26
- 12Auburn+8.23
- 13Mississippi State+8.03
- 14Texas A&M+6.76
- 15Vanderbilt-1.92
- 16Arkansas-2.55
Big Ten — 18 teams
- 1Ohio State+19.11
- 2Indiana+15.59
- 3UCLA+14.11
- 4Oregon+12.88
- 5Nebraska+10.06
- 6Northwestern+8.75
- 7Wisconsin+8.63
- 8USC+8.17
- 9Iowa+7.00
- 10Penn State+4.03
- 11Michigan+3.08
- 12Purdue+1.70
- 13Illinois+1.28
- 14Washington+0.87
- 15Michigan State-0.79
- 16Minnesota-0.96
- 17Maryland-2.93
- 18Rutgers-3.70
Big 12 — 16 teams
- 1Houston+13.97
- 2Utah+13.51
- 3Texas Tech+12.64
- 4BYU+11.51
- 5Baylor+8.97
- 6Arizona+7.57
- 7Oklahoma State+7.08
- 8UCF+5.92
- 9TCU+4.44
- 10Kansas State+4.05
- 11Iowa State+2.37
- 12Kansas+0.99
- 13Cincinnati-0.11
- 14West Virginia-1.45
- 15Colorado-2.69
- 16Arizona State-3.13
ACC — 17 teams
- 1Miami+19.33
- 2Wake Forest+10.12
- 3Louisville+8.19
- 4North Carolina+5.63
- 5Pittsburgh+5.52
- 6Duke+5.38
- 7Clemson+4.91
- 8Florida State+3.68
- 9Virginia Tech+3.23
- 10SMU+3.21
- 11Syracuse+0.21
- 12Virginia-0.00
- 13California-1.14
- 14NC State-1.97
- 15Georgia Tech-2.98
- 16Boston College-5.11
- 17Stanford-8.97
Pac-12 — 8 teams
- 1Boise State+8.00
- 2Oregon State+6.20
- 3Texas State+2.05
- 4Fresno State+0.80
- 5San Diego State-6.61
- 6Washington State-7.22
- 7Utah State-7.88
- 8Colorado State-9.33
American Athletic — 14 teams
- 1North Texas+3.72
- 2Memphis+2.11
- 3Army+1.33
- 4UTSA+0.88
- 5Tulane-0.46
- 6South Florida-0.87
- 7Tulsa-3.18
- 8Navy-3.80
- 9Florida Atlantic-4.66
- 10East Carolina-4.90
- 11UAB-7.76
- 12Rice-8.37
- 13Temple-9.16
- 14Charlotte-14.81
Sun Belt — 14 teams
- 1James Madison+8.48
- 2Georgia State-0.39
- 3Louisiana Tech-0.78
- 4Georgia Southern-1.21
- 5Louisiana-1.65
- 6South Alabama-3.13
- 7Troy-5.25
- 8App State-6.83
- 9Arkansas State-6.99
- 10Coastal Carolina-9.21
- 11Old Dominion-9.37
- 12Southern Miss-10.18
- 13Marshall-13.71
- 14UL Monroe-18.78
Mountain West — 10 teams
- 1New Mexico+8.34
- 2North Dakota State+3.87
- 3UNLV-1.75
- 4Air Force-1.76
- 5Nevada-8.40
- 6San José State-11.00
- 7Wyoming-12.09
- 8Hawai'i-14.53
- 9UTEP-16.02
- 10Northern Illinois-18.72
Conference USA — 10 teams
- 1Liberty+2.53
- 2Florida International-3.82
- 3Jacksonville State-3.87
- 4Western Kentucky-7.18
- 5Kennesaw State-8.11
- 6New Mexico State-9.09
- 7Delaware-10.11
- 8Sam Houston-10.98
- 9Middle Tennessee-12.67
- 10Missouri State-13.01
Mid-American — 13 teams
- 1Western Michigan+0.10
- 2Ohio-3.34
- 3Toledo-4.33
- 4Miami (OH)-6.61
- 5Eastern Michigan-9.23
- 6Central Michigan-9.53
- 7Massachusetts-9.70
- 8Buffalo-10.25
- 9Kent State-12.29
- 10Bowling Green-13.40
- 11Ball State-13.64
- 12Akron-16.18
- 13Sacramento State-17.80
Methodology
Every constant below was measured on 3,256 out-of-sample games rather
than assumed. Full evidence lives in reports/BASELINE_2019_2025.md.
Opponent-adjusted scoring margin, solved iteratively. Home field is removed before rating, so a soft home schedule earns nothing. The 4.53-point average is then adjusted per game for elevation gain, travel, and eastward body-clock shift — centred so a typical game still gets 4.53. Frequent FCS opponents are rated as themselves instead of one pooled bucket.
Margins are compressed through cap · tanh(margin / cap) at cap 32.
36% of FBS games are decided by 28+ points — the NFL figure is nearer 8% — so an
unadjusted mean lets garbage time drive the ratings. Capping is worth about half
a point of MAE.
| Home field (average) | 4.53 pts |
| Blowout cap | 32 |
| Recency half-life | 12 wks |
| Margin SD | 24.2 pts |
Standardised importance from a leave-one-season-out fit on opponent-adjusted features. Success rate dominates.
| success rate | 0.5796 |
| explosiveness | 0.1533 |
| ppa per play | 0.1488 |
| stuff rate inverse | 0.1183 |
Allowed statistics enter inverted, so giving up less rates better.
| success rate allowed inverse | 0.5402 |
| explosiveness allowed inverse | 0.1889 |
| ppa allowed inverse | 0.1451 |
| stuff rate | 0.1258 |
Raw season efficiency is confounded by schedule, and CFB schedules barely overlap. Adjusting is worth 0.30 points of MAE overall and 0.70 in weeks 5–7.
It also revived a signal: raw explosiveness fitted at 0.016 and looked worthless. Opponent-adjusted it fits at 0.153. The feature was never weak — the measurement was confounded.
Margin asks who is better, so it uses feature differences. A total asks how much scoring the two teams generate together, so it uses sums plus pace — drives and plays per game. CFB tempo varies far more than the NFL’s.
| League-mean total | 13.6544 |
| Model total | 13.0446 |
| Market total | 12.5055 |
| Total residual SD | 16.36 |
The headline score is algebra on the independent margin and total, then the mean of 10,000 residual draws. The sportsbook line is shown next to it as the benchmark. On 3,256 games the independent MAE is still worse than the close (12.53 vs 12.16); this board publishes the disagreement anyway, because a market-anchored number cannot beat the market. Authority stays RESEARCH_ONLY.
Ratings and scoring priors start from last year, then fold in this year’s
scored games as they arrive — 14% of the blend per completed week, the
share measured on weeks 1–4. A final that CFBD has not yet flagged
completed still counts if both scores are on the box. Week 1 is
still last year only; by week 3 this year already carries 28% of the rating
and of the scoring prior, on top of opponent-adjusted efficiency from the
games already played.
The published score is the mean of 10,000 independent draws from the residual margin (SD 24.2) and total (SD 16.36). That is the sample size where the standard error of the mean (~0.24 pts) sits an order of magnitude below the model’s MAE, so more draws do not change a displayed score. The stream is seeded from the matchup, so a rebuild with the same inputs reprints the same board. Mean scores are the projection; a single draw is not.
A nested leave-one-season-out audit covered 3,702 games from 2021–2025. The market remains the better MAE forecast. Publishing it as the headline made every projection sit on the sportsbook line, which cannot produce an edge. The board now publishes the independent margin and total as the forecast, with DraftKings kept as the live benchmark. ATS on disagreements is still 51.11% with a CI that straddles 52.38% breakeven, so this is a challenger, not a cleared betting model.
Every feature is queried strictly before the week being forecast. CFBD’s
endWeek is a real filter, so a backtest cannot see its own answer.
Weeks 2–4 no longer treat one or two games of efficiency as a mature sample. The full efficiency estimate is blended at 30%, 50%, and 80% respectively, with the fitted preseason prior carrying the balance. On 778 games from 2021–2025, the blend lowered MAE from 13.65 to 13.22. Outcome-scale calibration in weeks 2–3 lowered the final transition MAE to 12.72; week 4 calibration was rejected because it made error worse.
Week 1 is preseason-only; weeks 2–4 are a separately measured transition, not the mature full matrix. Across weeks 1–6 the updated model MAE is 12.54 against the market’s 11.87. Treat the market as the better estimate early.






