Methodology

The philosophy

Most college football models cling too tightly to preseason expectations deep into November. The Wire's system is designed to aggressively trust current-season performance. By midseason, what a team did last year barely matters. By late October, the preseason prior is essentially dead.

The Wire publishes two rankings — one computed by the engine, one cast by hand:

Four components

1. Preseason rating

Every team starts with a preseason number built from a prior-year baseline and a roster grade. The roster grade weighs returning production (50%), incoming talent from recruiting and the transfer portal (35%), and a coaching modifier (15%).

Within returning production, the model weights quarterback and offensive line production most heavily (35% each), followed by receiver/tight end production (25%), with running back production at just 5%. On defense, sacks lead at 30%, followed by tackles for loss and defensive snaps at 25% each.

Transfer portal additions receive a 1.3x multiplier over incoming freshmen, reflecting that proven college players contribute more immediately. Portal losses are netted against additions, but only FBS-to-FBS transfers count as losses.

2. Information score

This is what makes the Wire different from other systems. Every game a team plays earns "information points" based on two factors: how good the opponent was, and how surprising the result was.

A game against a top-10 opponent earns 5.0 base points. Opponents ranked 11th through 25th earn 4.0, 26th through 50th earn 3.0, 51st through 80th earn 2.0, 81st through 110th earn 1.25, and everyone below that earns 0.75. A game against an FCS team earns 0.25. The opponent's rank is taken from the edition published before kickoff, so a game's information value is set the day it is played and never re-graded.

These base points are then multiplied by a surprise factor. A result exactly as expected earns 0.8x; the multiplier rises in a straight line to 1.4x for a result 18 or more points from the expected margin. The expected margin is the gap between the two teams' ratings on that pre-game edition, plus 2.5 points for the home team. Neutral sites get nothing.

As information points accumulate, the preseason prior dies. Its share of a team's rating is (1 − points ÷ 30), squared: at 10 points the prior is down to 44% influence, and at 30 or more it reaches zero. Each team decays at its own pace based on the quality of opponents faced — a team that played three top-25 opponents by Week 4 gets judged on merit faster than a team that played three FCS schools. Every published rating carries this number, shown on each team's card as the preseason-versus-performance split.

3. Performance rating

For every game, the engine measures five things about how a team played, each as the team minus its opponent: EPA per play (33%), success rate (24%), explosiveness (19%), finishing drives (19%) — points per trip inside the opponent's 40 — and skill-based turnovers (5%), where interceptions count in full and fumbles count half, because recoveries are mostly luck. Garbage time is excluded.

Each measure is standardized against every FBS-versus-FBS game played so far this season, combined with the weights above, and converted to points by fitting the composite to actual scoreboard margins across those same games. One standard deviation of efficiency is worth whatever it actually produced on scoreboards this season, not a number we picked. A game's performance margin is 75% that efficiency margin and 25% the scoreboard margin.

Margins are capped at 28 points beyond expectation. A result counts only up to 28 points more or less than the pre-game edition expected, in either direction, so a 56-point favorite winning by 53 reads as a game that went to form rather than as a 28-point performance.

A game's performance margin, adjusted for home field and added to the opponent's current rating, is what that game says the team's strength is. A team's performance rating is the average of those readings, weighted by each game's base information points, and the whole board is re-solved from scratch every week across all 138 FBS teams so every rating rests on every opponent's current rating. The league average is held at the preseason average. A team's WPR is then its prior share times its preseason rating plus its performance share times its performance rating.

4. Game-day adjustments

Home field is worth a flat 2.5 points in the engine's expected margins for now. Venue tiers (1–4 points), rest, travel, and game-type modifiers such as a rivalry compression are planned for the prediction tools and will be documented here before they ship. The Spread Picker is neutral-field only, which is the WPR's native question.

The Wire Poll ballot

Starting with the Week 4 Edition, the Wire Poll is a personal top 25: a ballot, not a formula. Each week's ballot is cast the way a poll voter fills one out, and it is published in full. Movement arrows compare each ballot with the one before it.

The retired résumé layer

Through the Week 3 Edition, the Wire Poll was computed: the WPR plus 1.5 points per top-25 win and 0.75 per top-50 win, minus 0.5 per loss to a top-10 team and 3.5 per loss to a team ranked below 50th, with each opponent judged on the edition published before kickoff. Those editions stay in the archive as printed, and the layer stays in the engine so they can still be checked.

Transparency commitment

Every methodology change made during the season will be documented on the blog before the next set of rankings is published. If a weight changes, if a formula is adjusted, if a data source is swapped — it will be explained publicly.