Designed to improve as the season develops

The model is intentionally built to become more informative as more games are played. Early in a season, teams have small sample sizes, schedules vary dramatically, and there is limited information available about the true strength of opponents.

Because of that, early-season rankings can sometimes look unusual or change significantly from one week to the next. That is a natural result of a data-driven model working with limited early-season information rather than carrying forward assumptions about where teams are expected to finish.

As the season progresses, the model gains more evidence. Strength of schedule becomes clearer, quality wins become more meaningful, losses can be evaluated in better context, and the relationships between teams become increasingly established. The rankings are therefore intended to become more representative and more reliable each week.

Preseason projection

Before games are played, there is no current-season résumé to evaluate. The preseason projection therefore estimates team strength using objective roster and performance information rather than assigning teams subjectively.

The projection begins with a decayed three-year measure of recent program strength, weighted most heavily toward the immediately preceding season. It then adjusts that baseline with returning production, current roster quality, quarterback outlook, and transfer-portal movement. The public predictive rating is reported on a 0–100 scale.

For historical projections, the model is designed to use only information that would have been available before that season began. During Weeks 1–3, the preseason signal is gradually replaced by current-season results. From Week 4 onward, preseason information is removed and the rankings are based entirely on the normal in-season points model.

Core inputs

• Wins and losses, including additional treatment for losses to FCS opponents.
• Road wins and home losses.
• Strength of schedule.
• Quality wins based on the system’s own rankings.
• Conference championship outcomes when they fall within the selected week range.

Performance adjustments

• Game-control scoring that rewards stronger wins.
• Bad-loss penalties for poor losses.
• Close-loss credit in defined situations.
• Bonuses for undefeated teams and top-10 offense or defense.

Weekly ranking snapshots

When a season and week are selected, the model evaluates completed games through that point and produces a snapshot of the information available at that time. Every FBS team is evaluated under the same scoring rules and ordered by its final system score.

The purpose is not to predict which teams will eventually be the best. It is to evaluate what each team has accomplished through the selected point in the season.

Data source

Underlying team and game data is sourced from CollegeFootballData.com. That data source does not determine, endorse, or sponsor the CFB Top 25 rankings.