Current-season performance
Completed games can contribute scoring margin, opponent quality, schedule strength, efficiency, explosiveness, possession, pressure, discipline, and other selected factors.
MODEL METHODOLOGY
How My CFB turns football factors into rankings
The model is designed for fans who want to choose what matters and still see how every result was produced.
01 · INPUTS
Football evidence, saved before it is used
Preseason context
Talent, returning production, prior performance, recruiting, and transfer evidence can provide context before teams have played enough current-season games.
Point-in-time snapshots
Every ranking and prediction references a saved snapshot and cutoff. Later corrections do not silently rewrite the evidence behind an older result.
02 · RANKINGS
The factors are normalized before they are combined
Your importance settings
Each enabled factor receives the importance selected in the model builder. Relative importance matters; simply multiplying every setting by the same amount does not change the ranking.
Comparable scales
Raw measures use different units, so the published model version converts supported evidence to comparable team values before applying the selected weights.
Visible explanations
Ranking pages retain each factor’s value, weight, and contribution so users can inspect why a team received its position instead of relying on an unexplained score.
03 · PREDICTIONS
Matchup forecasts use only eligible pregame evidence
Strictly before kickoff
A forecast revision is eligible only when it was frozen before that game’s persisted kickoff. Started and completed games are never backfilled with a new prediction.
Independent of betting lines
The independent model does not use a sportsbook line as an input. A saved market line may be displayed later for comparison, but it remains separate evidence.
Experimental conversion
The current conversion from model difference to projected margin and win probability is an unvalidated beta policy. It is transparent and reproducible, but it is not established betting advice.
What anonymous visitors see
The public board uses one ownerless baseline with the published Standard starting mix. It is not trained from, averaged across, or otherwise derived from private user models. Logged-in users can create their own weighting choices, while public pages continue using only the public baseline’s persisted outputs.
Schedules, scores, and team evidence are refreshed offline. Opening a page never triggers a provider request, generates a ranking, or changes an analytical record.