Original model research · 2026
Where power and résumé disagree, and what the gap means
Compare scoring-strength ranks with wins above benchmark using the same completed schedule, with actual team examples and explicit limits on both measures.
668 classified teams have both ranks in this snapshot. The mean absolute gap is 12.0 places within a team's division; the largest is 98 places. These ranks answer different questions, so disagreement alone is not an error.
Stanford is a prominent FBS example: power No. 94, résumé No. 51, record 5-5, and −0.9 wins above benchmark.
- Mean absolute rank gap
- 12.0 places Each eligible team has equal weight; comparisons are within its own division.
- Teams compared
- 668 Only played, classified teams with a computed résumé; unclassified teams are excluded.
Strength and accomplishment use evidence differently
Power fits venue-adjusted scoring margins, with a soft cap, a win bonus, recency weights, and shrinkage. Résumé subtracts the wins a benchmark team would expect against a schedule from the team's actual wins, counting a tie as half a win. A narrow win and a rout both add one actual win to résumé, while their power contributions differ.
Résumé is not independent of margin. Its opponent ratings and benchmark strength come from the power model, which uses scores. The distinction is where margin enters: a team's own wins are counted as outcomes, while the difficulty of obtaining them is estimated through those ratings and venue probabilities.
Reproducing the benchmark
Within each division, the benchmark rank is min(25, ceil(ranked teams / 4)). With at least 97 ranked teams that is No. 25; a smaller dataset uses roughly its top-quarter team. For each opponent, calculate the benchmark's win probability at the same venue and sum those probabilities. The team's actual wins, with half credit for ties, minus that sum is wins above benchmark.
This study calls the same buildResumes function as the site, using the supplied published ratings and the snapshot's calibration, or the documented default calibration when insights are absent. It does not fit a new résumé formula to make the two lists agree.
Read the gap without inventing a cause
A positive gap in the example table means the power rank is better than the résumé rank. The team may have large scoring margins without as many difficult wins. A negative gap means résumé ranks it better than power does; close victories can contribute to that pattern. Neither explanation is established by the rank gap alone. The table adds record, average raw margin, and benchmark wins so readers can inspect the evidence.
The largest-gap table shows FBS when any eligible FBS teams exist; otherwise it shows the available divisions. It is intentionally selected for disagreement. The division summary gives the broader picture, including how many teams appear in both top-25 sets. An early sample or a fragmented schedule can make either ordering unstable.
Results from this snapshot
| Division | Teams | Mean absolute gap | Largest gap | Shared top 25 |
|---|---|---|---|---|
| FBS | 138 | 10.2 | 43 | 19/25 |
| FCS | 128 | 9.0 | 31 | 22/25 |
| Division II | 160 | 11.1 | 54 | 23/25 |
| Division III | 242 | 15.2 | 98 | 22/25 |
| Team | Power | Résumé | Gap | Record | Avg. margin | Benchmark wins | Wins above benchmark |
|---|---|---|---|---|---|---|---|
| Stanford | 94 | 51 | −43.0 | 5-5 | −13.5 | 5.88 | −0.9 |
| Kansas | 60 | 94 | +34.0 | 4-6 | +6.5 | 6.71 | −2.7 |
| UTSA | 54 | 23 | −31.0 | 8-2 | +4.8 | 7.42 | +0.6 |
| UConn | 71 | 101 | +30.0 | 5-5 | +8.0 | 8.06 | −3.1 |
| Nebraska | 21 | 50 | +29.0 | 6-4 | +15.4 | 6.84 | −0.8 |
| TCU | 43 | 72 | +29.0 | 5-5 | +8.2 | 6.69 | −1.7 |
| Wyoming | 89 | 60 | −29.0 | 7-3 | +1.4 | 8.33 | −1.3 |
| Ohio | 93 | 65 | −28.0 | 7-3 | +5.3 | 8.45 | −1.4 |
| Oregon State | 52 | 79 | +27.0 | 5-5 | +9.5 | 6.85 | −1.9 |
| Fresno State | 61 | 88 | +27.0 | 5-5 | +5.4 | 7.36 | −2.4 |
Inspect the team evidence
- Stanford: Power No. 94, résumé No. 51: 5-5 against a schedule yielding 5.88 expected benchmark wins; average raw margin −13.5.
- Kansas: Power No. 60, résumé No. 94: 4-6 against a schedule yielding 6.71 expected benchmark wins; average raw margin +6.5.
- UTSA: Power No. 54, résumé No. 23: 8-2 against a schedule yielding 7.42 expected benchmark wins; average raw margin +4.8.
Team profiles show the current publication. Their ratings can change after the snapshot used for this study.
Reproduce the study
- Use the provided published ratings for this same game set. Call buildResumes(teams, games, ratings, insights?.calibration ?? DEFAULT_CALIBRATION). The methods intentionally match the publication, including rounded display inputs when supplied.
- Match each played, classified team's division power rank to its computed résumé rank. Define the signed gap as résumé rank minus power rank; aggregate absolute gaps equally across teams.
- Compare within-division top-25 membership, using the smaller available population if necessary. Sort illustrative teams by absolute gap, then team id; do not treat those selected examples as a random sample.
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 v1-805b352f. The season and reporting periods identify the data studied; they are not publication or human-review dates.
Limitations
- This is a retrospective description of accomplishment and strength, not a prediction accuracy test or a reconstruction of the selection committee's choices.
- Résumé uses current opponent ratings and current calibration. Later results or source corrections can change how an earlier win is valued. Quality-win and bad-loss labels elsewhere are also retrospective unless explicitly labeled pregame.
- The two measures share model assumptions and source data; agreement is not independent confirmation. Disconnected division scales remain assumption-dependent.
- Teams with fewer results have a smaller opportunity to add wins above benchmark and more power shrinkage. This study reports no uncertainty interval for their rank gaps.
Found an incorrect score, calculation, or explanation? Send the page URL and supporting evidence. Publishing standards describe review and corrections.