HOW TO READ THIS SITE

Every rating on this site can be taken apart. A team's number is the sum of a season seed and one line per game it has played, and those lines add up to the rating exactly — we call that the receipt, and the site checks the sum before it publishes.

The Board ranks all 32 teams by that rating. Tap any team to see the receipt behind it.

The Board's rating — the Team Elo — works like chess Elo: before each game the model expects a result from the two ratings (plus a small home edge), then moves both teams by a fixed multiple of how far the actual result landed from that expectation. Beating a strong team by a lot moves you more than edging a weak one, and the scale is set so that one point of margin is worth about 15.6 rating points. Blowouts saturate — a 40-point win counts about the same as a 30-point one.

Ratings are walk-forward: a rating on any given week uses only the games played before it. That is why the time machine on the Board is exact at every week rather than an interpolation, and why the receipts still sum when you scrub back.

Neither model is simply better — it depends on the yardstick. On margins, the Ledger's measure, Team Elo's spreads usually sit a touch closer to the results; on win probability, the Factor model has been the slightly sharper of the two, most seasons. The Ledger keeps that score in public either way.

But no sport hangs on one player the way the NFL hangs on the quarterback, and a rating built from margins cannot see an injury — it finds out when the losses arrive, weeks late. That is what the second model is for: it rates the quarterback as his own component inside the offense, so when a starter is ruled out the offense re-prices before a snap is played. Most weeks the two models agree and the extra machinery looks redundant. The weeks they disagree are usually the weeks a quarterback is out — and those are exactly the weeks a single-number rating is most wrong.

A second model — the Factor Elos — rates each team's offense and defense in points per possession. Offense and defense are the two factors; the quarterback is rated as a component inside the offense. The Board's Off, QB and Def columns come from it. The team page itemizes the offense: supporting cast + quarterback effect + situation = offense. Cast and quarterback are separated by his individual per-dropback record and by what happens when someone else starts — not by team results alone.

The quarterback's own rating is portable — it travels with the player, seeded each season from his own record. The QB effect is how much he moves this particular offense above or below its cast. The half-and-half weighting of quarterback and cast is the model's design choice, held fixed — the data cannot pin that split down precisely, and the site does not pretend otherwise.

Situation is the odd one out. Teams starting someone other than their established quarterback underperform their parts by a few points — while quarterbacks with the same rating profile show no such miss when starting as the established man. The pattern is consistent with the team's disrupted phase, not the replacement's quality, so the model keeps it as a separate line that fades over his first starts instead of blaming the player.

Team Elo only reads scoreboards. It does not know who is starting, so a quarterback injury reaches it slowly, one result at a time. The Factor Elos price the announced starter before kickoff: the backup's rating replaces the starter's inside the offense, and the situation penalty is applied on top.

So when a starter goes down, the Factor spread moves that same week while Team Elo catches up over the games that follow — and the reverse when he returns. A gap between the two spreads on a game card is usually saying exactly that; the quarterbacks listed on the card are the first place to look.

Describing teams and pricing games honestly. Both models have been tested against the closing betting line and neither adds anything the line does not already know — the NFL is the most efficient sports market there is. Where the site shows a model spread next to a betting line, read it as a comparison of two opinions, one of which has real money behind it.

Elsewhere on the site: This Week (every game with its win probability and what it moves), Playoffs (the real seeding rules, simulated), the Ledger (both models' spreads beside the line, scored every week), Quarterbacks, and Compare (any two teams, side by side).