Preseason rating = last season's SP+ regressed toward the mean, nudged by returning production, recruited talent and recruiting-class strength — the weights LEARNED from 10 seasons by walk-forward, not guessed. Win totals are a Poisson-binomial over the real schedule; game randomness and rating uncertainty are both folded in.
Returning-production and preseason SP+/FPI for 2026 publish in late summer. Until then this regresses 2025 final ratings toward the mean — it will sharpen automatically when those inputs land. Honest by design: every number ships with a range, never false precision.
We tested recruiting-class strength and transfer-portal movement on top of the base (last-year rating + returning production + talent). Recruiting IMPROVED the backtest (8.06→7.90 rating RMSE) and is included; the transfer portal did NOT add signal beyond talent + recruiting (8.07 RMSE) and is honestly left out.
| Feature set tested | Corr | RMSE |
|---|---|---|
| base (carry+ret+talent) | 0.785 | 8.06 |
| + recruiting ✓ chosen | 0.793 | 7.90 |
| + portal | 0.784 | 8.07 |
| + recruiting + portal | 0.792 | 7.91 |
| # | Team | Outlook | Rating | Projected wins | Proj | '25 SP+ |
|---|---|---|---|---|---|---|
| 69 | Western Kentucky | Bowl team | -1.6±7.9 | 6 9 | 7.4 | +1.6 |
| 77 | Louisiana Tech | Bowl team | -3.4±7.9 | 5 9 | 7 | -1.3 |
| 89 | Kennesaw State | Bowl team | -6±7.9 | 5 9 | 7 | -5.4 |
| 92 | Jacksonville State | Bowl team | -6.8±7.9 | 6 9 | 7.3 | -6.7 |
| 102 | Liberty | Bowl team | -8.2±7.9 | 5 8 | 6.7 | -9 |
| 107 | Florida International | Bubble | -9.1±7.9 | 5 8 | 6.3 | -10.5 |
| 108 | Missouri State | Bubble | -9.2±7.9 | 4 7 | 5.6 | -10.7 |
| 109 | Delaware | Bubble | -9.4±7.9 | 4 7 | 5.6 | -10.9 |
| 123 | New Mexico State | Bubble | -12.2±7.9 | 4 7 | 5.2 | -15.5 |
| 126 | Middle Tennessee | Bubble | -12.5±7.9 | 4 7 | 5.3 | -16 |
| 129 | UTEP | Rebuild | -13.5±7.9 | 3 7 | 5 | -17.5 |
| 135 | Sam Houston | Rebuild | -19.9±7.9 | 2 5 | 3.4 | -27.8 |
Method: next-season rating is last year's SP+ regressed toward the mean (carryover 62%, learned from 2019+), plus returning production and recruited talent when published. Win totals are a Poisson-binomial over the real 2026 schedule with game randomness and rating uncertainty both folded in — so the range is the believable band of outcomes, not a guarantee. Backtest is strictly walk-forward (fit on prior seasons only). This is a season-level outlook, not a per-game spread. Source: CollegeFootballData (SP+, returning production, talent, schedule).