CFB RANKINGSStart on equal footing.
Unverified roster priors are disabled. Predictive strength comes from opponent-adjusted completed-game performance. Early rankings carry small-sample labels.
COLLEGE FOOTBALL / OPEN METHODOLOGY
Every game
earns it.
Completed games write the season. Every school starts on equal footing. See exactly what moves a team up.
Display filters never change model inputs or national ranks.
THE SEASON SHIFT
From potential to proof.
The published blend moves toward achievement as teams complete games.
Predictive —Achievement —
Published weighting
Advanced blend experiment
Change the share of record difficulty within achievement. Local experiment only; predictive component stays fixed. The remaining achievement share follows this edition’s quality/SOS/performance split; bounded adjustments remain active. Head-to-head overrides apply only to published order.
THE BOARD
Ranked by the numbers.
| Rank | Team | Record | Blend | Predictive | Achievement | Stat coverage |
|---|
No teams match your search.
Scores are relative indices, not win probabilities. Published order can override score order for adjacent head-to-head winners. Select a team for the evidence.
EVERY MATCHUP
College football schedule
All kickoff times are Central (CT).
What if these teams played?
Team versus team
Compare two teams at home, away or on a neutral field. This calculator runs locally and makes no paid AI calls.
THE STAT EXPLORER
Put two numbers together.
Compare teams, quarterbacks, or offensive playmakers. Hover, focus or tap a point for its identity. Double-click or press Enter for its team; team links in the table work on touch screens. Missing values are omitted.
Schedule strength and opponent-adjusted performance
| Team | Record | SOS | SOS score | SOR | SOR score | Raw offense | Adjusted offense | Raw allowed | Adjusted allowed | Covered FBS games |
|---|
Axes & display
Player qualification
Usage means pass attempts for quarterbacks, touches for offense and recorded tackles/disruptive plays for defense—not snaps.
Export image
Default export is exactly the graph, including its axes and quadrant labels. Logos never include a key.
| Name | X | Y | Sample |
|---|
AROUND THE COUNTRY
Conference comparison.
| Conference | Teams | Top 25 | % of conference ranked | % of Top 25 | Highest | Lowest | Average rank | Average score | Win rate | Average SOS rank | Average SOS score | Average SOR rank | Average SOR score | Adjusted points / game | Adjusted allowed / game | SOS/SOR teams covered |
|---|
Title-game outlook
Simulated title-game chances using current strength. Matchups are approximate: full eligibility, division and conference tiebreaker rules are not implemented.
| Team | Record | Expected wins | Title game | Champion |
|---|
A LOOK AHEAD
Your field. Your format.
First five out
Projected final top 25
Forecast assumptions
OPEN THE MODEL
One published model. Room to experiment.
The season slider changes only this browser. The owner can adjust every main-model weight through the versioned configuration and review changes before a manual release.
Sage · Advanced Reasoning Ranking Model
Sage is our learned statistical ranking model. It learns which game statistics predict results across seasons, then ranks teams by their average neutral-field win probability against the FBS field. It is separate from the main ranking blend and is not an LLM voting on teams.
Inspect the latest training candidate and validation metrics
| Rank | Team | Neutral-field average win probability |
|---|
Learning objective and coefficients
Training learns generic statistical relationships across years. Each team starts a new season with new evidence. A candidate needs owner approval before becoming the active model. No school or conference names are learned features.
NO HIDDEN BALLOT
A ranking you can inspect.
Measure each final.
Scoring margin, PPA per play, and success rate form the base performance signal. Supporting box-score evidence contributes up to 15% by default: yards, penalties, team points per possession minute, third/fourth-down conversion and quarter control. Missing detail weight stays with the base. Base detail uses cumulative leads through Q3; the achievement control proxy uses Q1–Q4 endpoints. Neither measures exact time leading. Advanced stats exclude provider-defined garbage time; box totals do not.
Reward the season.
SOR, quality results, measured performance and our continuous SOS score form achievement. Achievement reaches 100% after week-seven results. There is no conference multiplier.
Read the assumptions and current limitations
This is an experimental model, not a proven improvement over AP. Published weights and probability scales require historical calibration. Conference names are display labels only; no AP, CFP, recruiting stars, or provider team ratings enter the calculation.
Personnel priors are disabled because roster membership does not establish a current role. Offensive-line grades, defensive coverage grades, snaps, injuries, and depth charts are not available in this implementation. No chemistry penalty is invented.
Lower-division opponents use a division baseline learned from direct FBS games, their record, and one level of opponent-record context. It never recursively rates D-II, D-III, or NAIA teams. Missing records use a neutral estimate.
Achievement is 100% of the blend after week-seven results (Sunday 8 AM Central). Default achievement components are 60% SOR difficulty, 20% quality wins, 15% measured performance and 5% SOS score, plus bounded expectation penalties and opponent-weighted streak adjustments. Available median provider spreads can contribute to the penalty; they are not certified closing lines and may reflect market opinion. The performance component uses recent opponent-adjusted game performance, quarter-end control and available passing/rushing efficiency on offense and defense. Quarter-end control is not exact time leading. Missing supplemental coverage retains base performance weight. Adjacent head-to-head winners move ahead unless their series is split or part of a cycle.
When advanced data arrives late, final scores count immediately; available components are reweighted and coverage is shown. Later corrections update live statistics. Rankings publish once on Sunday at 8 AM Central and remain fixed until the next edition.
Settings for this edition
Data notes
Teams & season predictions
Expected wins sum game probabilities. Predicted winners are individual picks, not guaranteed outcomes.
| Team | Conference | Record | Expected wins | Playoff estimate | Conference final | Conference champion |
|---|
Postseason probability report
Model-based scenarios, not a forecast of committee votes. Conference tiebreakers are approximate. Six wins do not certify bowl eligibility.
| Team | Playoff | Conference title game | Conference champion | Six wins |
|---|
What would need to happen?
Find a concrete simulated path under the displayed model assumptions. Completed games cannot change. A failed search is not proof of elimination; full conference tiebreakers are not yet supported.
Bowl outlook
Named bowl assignments will appear only with verified selection contracts and eligibility inputs. See the six-win forecast in the probability report above.
Our prediction scorecard
| Probability band | Games | Average predicted win rate | Actual win rate |
|---|
Statistics, explained
Help improve CFB Rankings
Private feedback goes to the owner. Optional contact information is used only for follow-up. Do not include passwords or sensitive information. Review batches become ready after two weeks or 20 submissions.
Hypothetical moneyline results
This section is intended for adults 21+. This acknowledgment is a site policy, not verification of your age or local wagering eligibility. No wagers can be placed here.
Return to rankings$100 hypothetical stake on the model’s winner in every eligible FBS–FBS game. Capture starts at 10 AM Central on game day. Failed or incomplete captures retry every 15 minutes before kickoff, using the first recorded forecast and the price observed on that attempt. Late retries are labeled with their actual time. Missing, closed, started, or stale-forecast games are skipped. Captured prices are not guaranteed executable. No retrospective picks.
This is a model audit, not financial advice or a recommendation to gamble. Real bets can lose money. Find confidential gambling support. No sportsbook affiliation.