CF-PWR: a PairWise-style ranking for FBS football

Through games of Nov. 4, 2026, CF-PWR ranks — first, — second and — third. Tuned on five seasons of results (2021–2025), its selection-day ranking picked the winner of — of — College Football Playoff games; the committee's own ranking picked —.

CF-PWR top 40 after Week 10

Rank is the number of head-to-head comparisons a team wins against the other 59 teams in the comparison pool (hockey PWR rules). Two teams tied on comparisons are ordered by their own comparison; three or more tied are ordered by winning percentage, then the CF-PWR index. AP is the Nov. 1 poll, CFP the committee's first ranking (Nov. 3), ESPN FPI and SOR from ESPN's résumé page (updated Nov. 4; ESPN columns cover the 30 teams that page lists). NR = not ranked / not listed. Click a column to sort.

Where each ranking puts every team

One row per team ranked by anyone, ordered by CF-PWR. The line spans the gap between the most and least favorable ranking; marker shape tells you whose ranking it is. Hover a row for exact ranks.

CF-PWR AP poll CFP committee ESPN Strength of Record

CF-PWR likes more than the polls do

Biggest gaps vs. the CFP committee's ranking (or AP where the committee doesn't rank the team). Ranks shown as CF-PWR → committee/AP.

CF-PWR likes less than the polls do

Same comparison, the other direction.

How close are the top 25s?

How many of each source's top 25 are also in CF-PWR's top 25, and the rank correlation (1 = identical order) for those teams.

Ranking	Shared of 25	Rank correlation

Why the rankings differ

AP Top 25 — opinion poll

About 60–70 sportswriters each submit a ballot. There's no required method, so voters lean on the preseason poll as an anchor, tend to drop a team for when it loses rather than who it lost to, and rarely move teams that keep winning. CF-PWR has no preseason anchor: week 1 counts the same as week 10.

CFP committee — a room of 13 people

The committee has written criteria (championships, strength of schedule, head-to-head, common opponents) but no formula, and it weighs "game control" and the eye test. CF-PWR uses the same four criteria, but spelled out as points in a head-to-head comparison so anyone can audit why Team A is above Team B.

ESPN FPI — a forecast, not a résumé

FPI predicts future games using efficiency, returning starters, recruiting and coaching history, so a 6-2 team with a top roster can sit above an 8-0 team. CF-PWR only knows who played whom, where, and the score.

ESPN Strength of Record — closest cousin

SOR asks how hard it would be for an average top-25 team to post the same record against the same schedule. It's pure résumé (margin ignored). CF-PWR blends that kind of record-and-schedule measure 50/50 with a capped margin rating, because the five-year backtest showed margin adds real predictive information.

Playoff games — CF-PWR picked the winner

—

of — games, 2021–2025, using only selection-day results

Playoff games — committee's ranking picked the winner

—

higher committee rank = predicted winner

Late-season + playoff games predicted by CF-PWR

—

current rules: 50% margin, head-to-head tiebreaks

Same test, record alone

—

"just count the wins"

Which formula best predicts what happened next?

Each season, ratings built from games through Week 9 predict every FBS-vs-FBS game from Week 10 through championship weekend; ratings frozen on selection day predict the playoff games. Non-playoff bowls are excluded because of opt-outs. Higher-ranked team = predicted winner. About 1,400 games total. Hatched bars are CF-PWR: the earlier 75%-margin version and the current one.

What the lookback changed in the formula

Strength of schedule needs depth. Hockey's heavy weight on opponents' opponents (54%) beat the basketball-style 25/50/25 split. A team's direct opponents' records are easy to pad; their opponents' records aren't.
Margin matters, capped. Adding a margin rating (blowouts capped at 4 touchdowns) was the single biggest improvement; the backtest's best mix was 75% margin. CF-PWR deliberately uses 50% instead, so record and head-to-head results count for more. That costs about one point of accuracy (see the bars), a trade made on purpose: a ranking that hands out playoff spots should reward winning, not just dominating.
Head-to-head breaks ties. When two teams split the comparison points, the team that won the game between them wins the comparison. The backtest is indifferent to this rule (within a game or two), so it's a choice of principle.
Things that didn't help and were left at zero: hockey's home/road win weighting and a "quality win" bonus. Both were already captured by schedule strength and margin.
Bad losses do. A small penalty for losing to teams outside the top 80 survived the backtest.
FCS games are discounted: an FCS opponent counts as a team with a — win percentage.

Selection day, replayed: committee field vs. CF-PWR field

For 2021–2023 (4-team playoff), CF-PWR's top 4. For 2024–2025, the 12-team rules: five highest-ranked conference champions plus the seven best remaining teams, using CF-PWR's order. Markers: ●● in both fields, ◐ committee only, ◑ CF-PWR only, ○ neither (bubble team).

Season	Team	W-L	CF-PWR rank	Committee rank	Selected by

What the replay says

2021: CF-PWR would have taken Ohio State over Cincinnati. Cincinnati lost its semifinal by three touchdowns.
2022: identical fields.
2023: the famous Florida State snub. CF-PWR wouldn't have taken 13-0 FSU either (5th, same as the committee). It would have taken Ohio State instead of Alabama, which lost its semifinal to Michigan.
2024: one swap. The committee took Tennessee (beaten soundly in round one); CF-PWR would have taken Miami.
2025: two swaps. CF-PWR had Notre Dame (left out by the committee) and BYU in place of Oklahoma and Alabama. Alabama beat Oklahoma in round one, then lost its quarterfinal 38-3.

Every playoff game, 2021–2025

Who each ranking said would win, using selection-day information only. ✓ = picked the winner, ✗ = didn't.

Season	Game	CF-PWR pick	Committee pick

The calculation, step by step

Win percentage (WP). Wins ÷ games, FCS games included.
Opponents' win percentage (OWP). Average record of everyone you played, excluding their games against you (so beating someone doesn't hurt your own SOS). Records count only FBS-vs-FBS games, so an opponent's FCS cupcakes don't inflate your schedule.
Opponents' opponents (OOWP). The average OWP of your opponents: the part of schedule strength that's hard to game.
RPI = —×WP + —×OWP + —×OOWP, minus a — penalty per bad loss (per game played).
Margin rating. A least-squares rating that explains every scoring margin given who played whom and where, with margins capped at — points so running it up earns nothing.
CF-PWR index = blend of the two on a common scale, — margin rating and the rest RPI. The RPI weights, margin cap and bad-loss penalty were picked by the five-year backtest; the margin share was set lower than the backtest's best on purpose, so winning counts for more than dominating.
Pairwise comparisons. The top — teams by index are compared one-on-one. In each pair a team earns a point for: higher index; better record vs. common opponents; each head-to-head win; better record vs. "teams under consideration" (top —). More points wins the comparison; a tie goes to the team that won the head-to-head game, then to the higher index.
Final rank = comparisons won, as in hockey's PWR. Two teams tied are ordered by their own comparison; three or more tied (which can form a rock-paper-scissors loop) by winning percentage, then index.

What it deliberately leaves out

Preseason rankings, recruiting, returning starters, brand names
Conference labels: a conference's strength shows up only through its teams' results
When a loss happened (an early loss counts the same as a late one)
Injuries, "eye test", style points beyond the margin cap

Known limits

Ten weeks in, one result can still move a team several spots; the pairwise layer is most stable after conference championships.
Pairwise rankings aren't always transitive: A can beat B, B beat C, and C beat A. Hockey lives with this; so does CF-PWR.
Thirty-one playoff games is a small sample. The broader test (about 1,400 late-season and playoff games; non-playoff bowls excluded because of opt-outs) is the one the weights were tuned on.
Data: game results from the open-source cfbfastR-data project (CollegeFootballData.com), plus the five MAC games of Nov. 3–4.

Head-to-head comparisons among the current top 16

Each cell is the row team's points vs. the column team's points in their pairwise comparison. Bold with ● = the row team wins that comparison (equal points go to the head-to-head winner, then the higher CF-PWR index). A bold cell below the diagonal is a lower-ranked team winning its comparison with a higher-ranked one: the non-transitivity pairwise systems accept.