Original model research · 2026
How much does the blowout cap change the rankings?
Re-solving the same completed games with 14-, 28-, and 42-point soft caps reveals which rankings depend most on large margins.
Across 3,347 completed games and 668 active teams, the largest within-division rank range across the three caps is 32 places. These are alternative descriptions of the same results, not forecasts.
926 games (28%) have a venue-adjusted margin above 28 points in magnitude. A soft cap changes smaller margins too: 28 is a curve parameter, not a cutoff that leaves every smaller score untouched.
- 14-point cap
- 3.8 places Mean absolute within-division rank change from the 28-point baseline, over all active teams.
- 42-point cap
- 1.9 places Mean absolute within-division rank change from the 28-point baseline, over all active teams.
The choice being tested
A 56-point win and a seven-point win should not tell the model the same thing. Whether they should differ by eight times is a separate modeling choice. The published model compresses venue-adjusted margins with C × tanh(margin / C), then adds the win bonus. Reducing C gives large victories less influence; increasing it preserves more of the scoring gap.
We change only C. Home field, win bonus, ridge shrinkage, recency, and cross-division weights stay fixed. This isolates sensitivity to the curve rather than comparing unrelated ranking systems. The alternative caps are stress tests, not settings chosen to produce a preferred team order.
Read rating changes and rank changes together
A smaller cap usually compresses rating differences. That can change a rating substantially without moving its rank. Conversely, a small point change can move a team past several near-equals. The comparison table reports both effects; the team table deliberately highlights the largest rank ranges, so it is not representative of every team.
The overlap column counts members retained from the default FBS top 25, using fewer teams when the dataset has fewer than 25 ranked FBS teams. It measures membership stability, not whether either list is correct. Follow a team link to inspect its scores and opponents before explaining a large change as dependence on blowouts.
Results from this snapshot
| Soft cap | Mean point change | Mean rank change | Largest rank change | FBS top-25 retained |
|---|---|---|---|---|
| 14 | 2.85 | 3.84 | 24 | 22/25 |
| 28 | 0.00 | 0.00 | 0 | 25/25 |
| 42 | 1.59 | 1.93 | 11 | 24/25 |
| Team | Division | Rank, cap 14 | Rank, cap 28 | Rank, cap 42 | Range |
|---|---|---|---|---|---|
| Olivet | Division III | 152 | 176 | 184 | 32 |
| Allegheny | Division III | 86 | 104 | 114 | 28 |
| Amherst | Division III | 67 | 83 | 90 | 23 |
| Lake Forest | Division III | 52 | 33 | 29 | 23 |
| Adrian | Division III | 74 | 58 | 51 | 23 |
| Monmouth (IL) | Division III | 58 | 43 | 35 | 23 |
| Huntingdon | Division III | 66 | 82 | 89 | 23 |
| MIT | Division III | 63 | 79 | 85 | 22 |
| Colby | Division III | 203 | 187 | 182 | 21 |
| Wittenberg | Division III | 189 | 178 | 168 | 21 |
| Raw margin | Cap 14 target | Cap 28 target | Cap 42 target |
|---|---|---|---|
| 7 | 8.97 | 9.36 | 9.44 |
| 28 | 16.00 | 23.82 | 26.98 |
| 56 | 16.49 | 29.49 | 39.04 |
Inspect the team evidence
- Olivet: Division ranks 152, 176, and 184 with caps 14, 28, and 42; a 32-place range.
- Allegheny: Division ranks 86, 104, and 114 with caps 14, 28, and 42; a 28-place range.
- Amherst: Division ranks 67, 83, and 90 with caps 14, 28, and 42; a 23-place range.
Team profiles show the current publication. Their ratings can change after the snapshot used for this study.
Reproduce the study
- Use the same validated, unique completed games and complete team dictionary as the published model; sort by season period and game id. Exclude unplayed teams from comparisons but keep them in each solve.
- Run rankTeams(teams, games, { ...DEFAULT_CONFIG, marginCap }) for marginCap 14, 28, and 42. Re-solve the 28 baseline to avoid rounding artifacts from published display ratings.
- Compute absolute changes from the 28 baseline for every active team. Average teams equally, not games. Compute each team's rank range as its maximum rank minus its minimum rank across all three solves.
The calculation uses 3,347 completed games and 668 active teams. 14 teams have no usable completed result.
Game results and team metadata come from College Football Data. CFB Power Board calculates the ratings, comparisons, tables, and commentary. The project source contains the research and rating functions named above. The model documentation explains the formula. Download the current source teams, source games, and published ratings and game logs. Filter source games to usable completed results for the study's season before running the methods above. These files are current snapshots, not a permanent historical archive.
Model settings used in this study
- homeFieldAdvantage
- 2.5
- marginCap
- 28
- winBonus
- 2.5
- ridge
- 1.2
- recencyDecay
- 0.97
- recencyFloor
- 0.6
- crossDivisionPenalty
- 0.55
- maxIterations
- 500
- tolerance
- 0.000001
- relaxation
- 1.5
Analysis specification 2026-11-07.1 · content revision v1-b3d5af8a. The season and reporting periods identify the data studied; they are not publication or human-review dates.
Limitations
- This is retrospective sensitivity, not an out-of-sample accuracy test. It cannot establish which cap predicts future games best.
- All three solves share the same data errors and assumptions. Stability across these caps does not validate the venue setting, shrinkage, division weights, or schedule coverage.
- A rank range is not a confidence interval. Cross-component and sparsely connected comparisons remain uncertain even if their ranks barely move.
Found an incorrect score, calculation, or explanation? Send the page URL and supporting evidence. Publishing standards describe review and corrections.