2016 College Football Rankings

Every other column on this page is bought in — Elo, SP+, FPI, SRS and the AP poll all arrive from CollegeFootballData. CORE is the exception. We build it here, from play-by-play, and it is published nowhere else.

It answers a narrower question than a power rating: how good is this team per play, once you remove the situations it happened to face and the opponents it happened to draw? Two teams can post identical yards per play when one of them ran its snaps at a three-score lead against the worst defenses on the schedule. That gap is most of the difference between the two numbers, and every raw per-play stat carries it.

So we remove both, in two stages. First the situation: each play’s expected value is predicted from game state alone — period, clock, score margin, down, distance, field position, home or away, pass or rush — with no team identity in the model at all, and what is left over is the part the situation does not explain. Then the opponent: offenses and defenses form one connected network over a season, so both sides are fitted at once by ridge regression rather than credited separately. Ridge specifically, because in September the schedule graph is barely connected and an unregularised fit hands a huge rating to whoever beat an overmatched opener.

Reported in points per 100 qualifying plays, above average. Offense is better when high, defense when low. Overtime, garbage time, FBS-vs-FCS games, special teams and non-scrimmage plays are all excluded.

⚠ CORE is a measurement, not a forecast. It says what a team has already done per play, adjusted. It does not say who wins next week — that is a separate model, and the two are kept as separate objects on purpose. Full method →