CFB Betting Edge

2025 · Week 3 · ACC · backtest — what the model predicted vs what happened, and vs the market
Week 3 backtest · ACC · 9 games — what the model predicted vs what happened
Straight up
6–367%
ATS vs close
6–367%
Model margin MAE
12.7
Market margin MAE
13.6
→ model's margins were closer than the market
Results — model read vs the final score + how the closing-line lean held up
Duke @ TulaneFinal 2734
Model
pred Tulane 13.3 · actual Tulane 7
winner Tulane
ATS vs close
leaned Tulane -1.5Cover ✓
margin err: model 6.3 · mkt 5.5 · mkt closer
Model
pred Miami 6.1 · actual Miami 37
winner Miami
ATS vs close
leaned South Florida +17.5Miss ✗
margin err: model 30.9 · mkt 19.5 · mkt closer
Model
pred Virginia Tech 5.1 · actual Old Dominion 19
winner Virginia Tech
ATS vs close
leaned Old Dominion +5.5Cover ✓
margin err: model 24.1 · mkt 24.5 · closer
Model
pred California 3.5 · actual California 13
winner California
ATS vs close
leaned California +3.0Cover ✓
margin err: model 9.5 · mkt 16.0 · closer
Model
pred Georgia Tech 0.7 · actual Georgia Tech 3
winner Georgia Tech
ATS vs close
leaned Georgia Tech +3.0Cover ✓
margin err: model 2.3 · mkt 6.0 · closer
Model
pred NC State 6.0 · actual NC State 10
winner NC State
ATS vs close
leaned Wake Forest +7.5Miss ✗
margin err: model 4.0 · mkt 2.5 · mkt closer
Model
pred Boston College 6.5 · actual Stanford 10
winner Boston College
ATS vs close
leaned Stanford +14.0Cover ✓
margin err: model 16.5 · mkt 24.0 · closer
Model
pred Pittsburgh 10.4 · actual West Virginia 7
winner Pittsburgh
ATS vs close
leaned Pittsburgh -7.0Miss ✗
margin err: model 17.4 · mkt 14.0 · mkt closer
Model
pred SMU 15.0 · actual SMU 18
winner SMU
ATS vs close
leaned Missouri State +28.5Cover ✓
margin err: model 3.0 · mkt 10.5 · closer
The model beats SP+ but not the closing line (ATS ≈ break-even). “Model−Mkt” is the gap between our projected home margin and the market's implied margin — a read on disagreement, not a profitable edge. Upcoming games have no closing line yet, so the market number marked ~ is the consensus midpoint of the best posted spreads across books.