🏈 Every NFL season since 1920

107seasons, each with every team’s rating week by week, its final standings and its best games. One model runs the whole way through, so a 1925 team and a 2025 team are measured the same way and 2026 is already seeded.

48,636 team-weeks · 19202026 · built 2026-09-06

Browse a season

107 seasons, one hub each. Hover a year for who ended it on top.

The belt: the best team in the league, season by season

One square per season, in that team's own colour. Runs of dominance are the point.

How this is measured
Highest rating at the end of the season, which is not the same as champion and disagrees 25 times. Reading across, a colour that holds for four squares is a dynasty and a wall of different colours is a decade nobody owned. A franchise with no stored colour keeps the neutral border rather than being assigned one.

A star marks a season where the best-rated team also won the championship. It happens in 81 of 106 finished seasons.

Against expectation, a century of it

Every game priced before kick-off, and scored afterwards against the closing line.

How this is measured
Brier is the squared error of a probability: give a team 0.8 and it wins, you are charged 0.04; give it 0.8 and it loses, you are charged 0.64. Lower is better, so this chart is drawn upside down and a higher line is a better forecast. The market is the number to beat and mostly it wins, which is the honest headline. Every season has its own priced game log.
this site’s modelthe betting markethigher is a better forecast

Tap or drag on the chart: the nearest season, the model's score and the market's.

The results a century of Elo got most wrong
beat New England Patriots, 2008
7%
beat Dallas Cowboys, 1995
8%
beat Washington Redskins, 1972
8%
beat Pittsburgh Steelers, 2002
8%
beat Los Angeles Rams, 2020
8%
beat Green Bay Packers, 1967
8%
beat New England Patriots, 2019
9%
beat Oakland Raiders, 2001
9%

The full method, the market comparison and the metro rollups are on Against Expectation. Each season’s own priced game log is linked from its season page.

When the best team did not win

25 of 106 finished seasons ended with the championship somewhere other than the top of the ratings.

How this is measured
The model does not know who won a title; it only knows results, margins and who they came against. So a gap between the best-rated team and the champion is not an error in either direction. It is the sport being a knockout at the end of a league, which is the whole reason anyone watches the end of it.

The years two leagues ran at once

Rated in one pool, so the comparison is finally possible.

How this is measured
The AAFC and the AFL each played alongside the NFL and each had its own table. Nobody could say how the leagues compared, because nobody rated them together. These seasons do.

1946, 1947, 1948, 1949, 1960, 1961, 1962, 1963, 1964, 1965, 1966, 1967, 1968, 1969. Every team in those years is rated against every other, whichever league it played in.

Where these numbers come from

Enough to disbelieve this page on purpose rather than by accident.

Ratings are Neil Paine’s NFL Elo, carried in this site’s own NFL workbook: 48,636 team-weeks across 107seasons, every one with a rating and a league rank. Championships and year-end honours come from the same workbook’s year-by-year record, not from the ratings.

Every league that ran in a season is rated in ONE pool, so the 1946-49 AAFC and the 1960-69 AFL are rated against the NFL of their day. Only the standings on a season page are split by league.

Nothing here is a forecast. The model was not held out from the seasons it rates, so read it as a description of what happened rather than a prediction of what will. Against Expectation is where the same ledger is scored against the betting market, and it loses. Built 2026-09-06.