LEILA Ratings Methodology

What the numbers actually mean

LEILA Ratings is built to describe team strength from real games, not recreate a poll. The core ratings account for opponent quality so the same statistical performance is treated differently against a strong opponent than against a weak one.

Core

Ratings

In-season Adj. Net, Adj. Off and Adj. Def use only games from the season being rated. No prior-year team rating, recruiting input or preseason ranking is blended into LEILA’s published ratings. Early-season estimates are stabilized only by regularization inside the current season’s opponent network.

Adj. Net

Core

Overall opponent-adjusted rating. Adj. Net = Adj. Off + Adj. Def

Adj. Off

Core

Opponent-adjusted offensive rating combining EPA, Success Rate, and Explosiveness. Higher is better.

Adj. Def

Core

Opponent-adjusted defensive rating combining EPA, Success Rate, and Explosiveness. Higher is better.

SOS

Context

Strength of schedule, based on the strength of opponents played through the selected rating snapshot.

SOR

Résumé

Strength of Record: wins above what an exactly-average FBS team would be expected to earn against the same schedule. It answers a résumé question, separate from Adj. Net's performance question.

ASM

Résumé

Adjusted Score Matrix: an opponent-adjusted scoring-margin rating (constrained least squares), with each game's margin capped at 28 points before fitting so a blowout can't dominate a team's number. A résumé lens like SOR, not a second opinion on who's better -- see “Process vs. résumé” below. Shown on Advanced, not the main Ratings page.

Reading The Ratings

Process vs. résumé

Adj. Net is LEILA’s primary answer to “how good is this team” — fit on per-play process (EPA, Success Rate, Explosiveness) rather than the scoreboard directly, and walk-forward tested against games it hadn’t seen to confirm it actually predicts better than a plain margin-based rating, not just fit one better. ASM will look deceptively strong on a same-season fit test, since it’s built to minimize error against exactly that season’s scores — but tests that strictly predict future, unseen games show Adj. Net calling more of them correctly, especially early in a season before a margin-only rating has enough games to stabilize. Read ASM as what the scoreboard says, with blowouts capped — most useful for where it disagrees with Adj. Net (a team over- or under-performing its underlying process is a real story), not as a competing opinion on which team is actually better.

Team Profiles

Raw vs adjusted

LEILA Ratings does not label every number opponent-adjusted. Public team profiles include season-to-date raw football results such as success rate, pass/rush success, yards per play, explosiveness and tendencies. Adj. Net and fields explicitly labeled as adjusted are model-based opponent-adjusted snapshots.

Advanced

Week ranges

Metrics built from additive weekly counts can be recalculated for a selected range. Model ratings are snapshots and cannot honestly be added or averaged across arbitrary weeks, so LEILA Ratings keeps those two classes separate instead of pretending every column is rangeable.

Schedule

Week 0

LEILA Ratings uses chronological site weeks. Early opener games that occur several days before the main Week 1 slate are separated into Week 0 so the interface follows how fans actually experience the season instead of forcing every early game into one label.

This Week

Pregame snapshots

Weekly matchup pages use the latest LEILA rating snapshot strictly before the selected game week. That prevents the matchup page from quietly using the result it is supposed to be previewing.

Model Research

Predictions must earn trust

Walk-forward testing. Historical test seasons are evaluated using only seasons and games available before the test sample.

Immutable live predictions. Prospective prediction snapshots are timestamped and are not silently regenerated after the game.

No fake probability precision. LEILA Ratings will not publish a model win probability as calibrated until calibration has actually been validated.

Misses stay visible. The long-term goal is a public prediction archive so users can inspect both correct calls and misses.

Use LEILA Ratings

Start with the football

Check the national ratings, then open the weekly slate to see how the teams in this weekend’s games compare before kickoff.

View ratings →Open Predictions →