🎲 Against Expectation
The record book says who won. It cannot say who was lucky, who was robbed, or which result was genuinely impossible. Every match in six European top flights since 1888 and every NFL game since 1920 is scored here against the chance it was given before kick-off, on one model, which turns out to reveal several things none of those leagues has ever announced.
Home advantage is dying, in both sports at once
Not a rule change, not a season, not one league.
How this is measured
English football spotted the home side 191 Elo points at its peak in 1901-02. Last season it was 56. The NFL went from 190 in 1925 to 29 in 2025, close enough to nothing that playing at home is now worth less than the gap between two mid-table sides.
The measure is Elo points rather than home-win share on purpose. About a quarter of English league matches are drawn and almost no NFL game is tied, so the two leagues’ home-win percentages are not comparable and never were. The rating gap a home side is effectively spotted is: 400·log₁₀(home wins / away wins), in which the draws divide out. That is the only reason these two lines are allowed to share an axis.
The longest odds ever beaten
Nobody curated these; the model went looking for improbability.
How this is measured
English top flight
| # | Match | Chance |
|---|---|---|
| 1 | Liverpool 1-2 Leicester CityLeicester City won away | chance the winner had 3.8% |
| 2 | Manchester United 0-1 Derby CountyDerby County won away | chance the winner had 4.5% |
| 3 | Liverpool 0-1 Wolverhampton WanderersWolverhampton Wanderers won away | chance the winner had 4.7% |
| 4 | Manchester United 2-3 Blackburn RoversBlackburn Rovers won away | chance the winner had 5.0% |
| 5 | Manchester United 0-1 West Ham UnitedWest Ham United won away | chance the winner had 5.4% |
| 6 | Manchester City 2-3 Crystal PalaceCrystal Palace won away | chance the winner had 5.5% |
| 7 | Leeds United 1-3 SouthamptonSouthampton won away | chance the winner had 5.5% |
| 8 | Chelsea 1-2 SunderlandSunderland won away | chance the winner had 5.7% |
| 9 | Liverpool 0-2 Norwich CityNorwich City won away | chance the winner had 5.8% |
| 10 | Manchester United 0-1 West Bromwich AlbionWest Bromwich Albion won away | chance the winner had 6.3% |
NFL
| # | Match | Chance |
|---|---|---|
| 1 | Miami Dolphins 38-13beat New England Patriots | chance the winner had 6.5% |
| 2 | Washington Redskins 24-17beat Dallas Cowboys | chance the winner had 7.5% |
| 3 | Buffalo Bills 24-17beat Washington Redskins | chance the winner had 7.6% |
| 4 | Houston Texans 24-6beat Pittsburgh Steelers | chance the winner had 7.9% |
| 5 | New York Jets 23-20beat Los Angeles Rams | chance the winner had 8.0% |
| 6 | Pittsburgh Steelers 24-17beat Green Bay Packers | chance the winner had 8.2% |
| 7 | Miami Dolphins 27-24beat New England Patriots | chance the winner had 8.7% |
| 8 | Arizona Cardinals 34-31beat Oakland Raiders | chance the winner had 9.0% |
| 9 | Houston Oilers 34-21beat Cincinnati Bengals | chance the winner had 9.1% |
| 10 | Carolina Panthers 13-7beat San Francisco 49ers | chance the winner had 9.2% |
The seasons that broke the model
The ratings still could not see these coming.
How this is measured
English top flight, league points vs expected
| # | Club | vs expected |
|---|---|---|
| 1 | 2015-16Leicester City | +27.4 |
| 2 | 1985-86West Ham United | +24.5 |
| 3 | 2016-17Chelsea | +22.3 |
| 4 | 2017-18Manchester City | +21.4 |
| 5 | 1989-90Aston Villa | +21.4 |
| 2007-08Derby County | −25.1 | |
| 1984-85Stoke City | −23.9 | |
| 2002-03Sunderland | −20.9 | |
| 2018-19Huddersfield Town | −19.3 | |
| 1987-88Watford | −18.8 |
Spain, Italy, Germany, France and the Netherlands, match points vs expected
| # | Club | vs expected |
|---|---|---|
| 1 | 1973-74Feyenoord | +9.21 |
| 2 | 2023-24Bayer Leverkusen | +8.68 |
| 3 | 2011-12Montpellier HSC | +8.01 |
| 4 | 2023-24VfB Stuttgart | +7.99 |
| 5 | 1988-89Inter | +7.97 |
| 1945-46FC Sochaux | −10.59 | |
| 2018-19AS Monaco | −9.12 | |
| 1965-66Tasmania 1900 Berlin | −8.73 | |
| 1994-95Brescia Calcio | −8.27 | |
| 1988-89RC Lens | −8.15 |
NFL, wins vs expected
| # | Team | vs expected |
|---|---|---|
| 1 | 2004Pittsburgh Steelers | +5.42 |
| 2 | 1999Indianapolis Colts | +4.94 |
| 3 | 1981San Francisco 49ers | +4.65 |
| 4 | 2025Jacksonville Jaguars | +4.53 |
| 5 | 2025New England Patriots | +4.52 |
| 1981New England Patriots | −5.68 | |
| 1994Houston Oilers | −5.45 | |
| 2013Houston Texans | −5.29 | |
| 2001Carolina Panthers | −5.25 | |
| 2014Tennessee Titans | −4.76 |
Six leagues, one model, and the most predictable league in Europe
Skill is how much the ratings knew that the era's home-and-away split did not.
How this is measured
| League | Skill |
|---|---|
| NetherlandsEredivisie | skill against that league's own era baseline 8.15% |
| ItalySerie A | skill against that league's own era baseline 5.65% |
| SpainPrimera División | skill against that league's own era baseline 4.75% |
| GermanyBundesliga | skill against that league's own era baseline 4.10% |
| EnglandEnglish top flight | skill against that league's own era baseline 3.17% |
| FranceLigue 1 | skill against that league's own era baseline 2.96% |
The Eredivisie is the most predictable of the six and the Ligue 1 the least. Predictable is not a judgement on the football. It means the same model, fitted the same way, learns more from the same number of matches, which usually points at a settled gap between a few clubs and the rest rather than at the quality of the play.
The number is the improvement on a baseline that knows nothing about the two clubs, only how often that era’s matches ended home, away or drawn. A league at 0% would be one where knowing who is playing tells you nothing you did not already know from the fixture list.
Nobody publishes this comparison, which is mostly because nobody runs one model across six leagues and a century. The settings were fitted on England and reused unchanged, so the transfer is a stated choice rather than a tuned result: skill is positive in all six, and refitting per league is available and not done.
By metro, and the thing size does not buy
The biggest metro is not the one that beats its odds.
How this is measured
English top flight, match points vs expected
| # | Metro | vs expected |
|---|---|---|
| 1 | Liverpool | +94.6 |
| 2 | Manchester | +75.8 |
| 3 | London | +50.2 |
| 4 | Newcastle | +13.4 |
| 5 | Bournemouth | +7.2 |
| 6 | Coventry | +6.4 |
| 7 | Brighton & Hove | +4.5 |
| 8 | Sheffield | −0.4 |
| Middlesbrough | −21.6 | |
| Nottingham | −22.8 | |
| Leicester | −23.2 | |
| Blackburn-Burnley | −26.6 |
The five continental top flights, match points vs expected
NFL, wins vs expected
| # | Metro | vs expected |
|---|---|---|
| 1 | Green Bay | +25.4 |
| 2 | Los Angeles | +23.0 |
| 3 | Dallas | +17.4 |
| 4 | Pittsburgh | +15.6 |
| 5 | Miami | +14.9 |
| 6 | Seattle | +13.7 |
| 7 | Cleveland | +11.8 |
| 8 | San Francisco-San Jose | +10.0 |
| Detroit | −13.2 | |
| Phoenix | −16.5 | |
| Tampa | −21.1 | |
| New York | −26.8 |
Madrid tops the continental board at +167.7 match points from 5 clubs. Rotterdam-The Hague has 16 clubs in these leagues, more than anywhere else, and sits at +28.1. Depth and dominance are different quantities and a league table cannot separate them. Totals are safe to compare within a league and only loosely across them, because league size and era differ.
Form against money
Squad value at the end of 2024-25 against that same season's surplus, for every club the two ledgers share.
How this is measured
Rayo Vallecano outperformed its money by the most in 2024-25: ranked 17th on squad value in the Primera División but 2nd on surplus, a gap of +15. Real Sociedad sits furthest the other way, a gap of −13 between its Primera División value rank (4) and its surplus rank (17).
Show all 82 clubs
Return on the transfer window
Fees paid and received from 2012-13 to 2025-26, and what each squad gained in value once the trading is netted out.
How this is measured
Ajax Amsterdam made the most of its windows: +€626m of squad value beyond what it paid, across 14 priced seasons. Chelsea FC spent the most, €2.88b, for −€977m. Manchester United sits at the foot, −€1.33b. The best return on a real programme of spending is AZ Alkmaar, +€341m on €99m spent.
Show all 192 clubs
Money against football
One dot per club and season: points against expectation across, squad value gained beyond the spend up.
How this is measured
6 club-seasons make the frontier. The furthest right is Leicester City in 2015-16, +9.5 points against expectation with +€28m of value gained beyond its spend; the highest is Liverpool FC in 2017-18, +€299m.
Tap or drag on the plot: the nearest club-season, its points and its money.
Grown or bought
Each club's squad-value change from 2021-22 to 2025-26, split into trading and appreciation.
How this is measured
VfB Stuttgart grew the most squad value over the window, +€352m of appreciation beside +€137m of trading. 1.FC Union Berlin grew it the purest way: 1764% of its value change was appreciation, not spending.
Show all 111 clubs
The director's ledger
Every arrival of the last five seasons graded on realised outcome, 2022-23 to 2026-27.
How this is measured
SPAL has the best multiplier on real spending, 39.1× on €0m deployed. The best single trade league-wide: FC Barcelona’s Lamine Yamal from Barça U19, €0m to +€200m (held). The worst: Paris Saint-Germain’s Randal Kolo Muani, −€75m.
Show all 204 clubs
Against the market, which usually wins
Both ledgers are graded against the betting market, and on balance both lose.
How this is measured
24 seasons priced, 9,116 matches. The model was closer in 1 of them: 2008-09.
Over those matches the model scores 0.5809 and the market 0.5709. Across the whole 1888 history the model’s Brier is 0.5964, a 3.2% improvement on knowing nothing but the era’s home-and-away split.
11,862 games head to head. The model was closer in 8 of 47 seasons.
Model Brier 0.2198, market 0.2134.
There is no third card, because there is no third market. The five continental ledgers carry no price at all, so Spain, Italy, Germany, France and the Netherlands are measured against the model and nothing else. A continental line on this section would be an implication the data cannot support.
Your own calls land on the same axis: Citizen of Nowhere Picks scores a pick with the identical Brier the 1888 and 1920 ledgers use, so a call made this weekend is measured against a century of them.
Where these numbers come from, and what is wrong with them
Everything a reader needs to disbelieve this page on purpose rather than by accident.
The model
A three-way Elo, one rating per club, updated after every match and scaled by the goal margin, with ratings pulled back toward the middle between seasons because promotion and relegation churn the pool every year. Home advantage and the draw rate are estimated from the five seasons before the one being priced, never the season itself, so every number the model used at kick-off was available before kick-off. The NFL ledger is the same idea on a pre-game win probability the workbook carries for all 106 seasons.
What it cannot do
Nothing is held out: the handful of settings were fitted on the whole history, so treat this as a description of the record rather than a forecast. And the football model has almost no skill before about 1960: knowing the era’s home-and-away split is nearly as good as knowing which two clubs are playing. It earns its keep from roughly the 1960s on, and most of it in the last twenty-five years.
How it is checked
Every season’s league table is rebuilt from the matches and compared against the table this site already publishes on its own season hubs: 67 seasons, 1,405 club-seasons, 0 unmatched club names. Separately, every club’s count of top-flight seasons is compared against the number on its own club page, and all of them agree. That second check is what caught 1939-40: the league ran three matchdays before war was declared and the results were expunged, but the source still carried the matches, so a phantom season was appearing on every club that played them.
What is still wrong
The checks leave 2 fixtures in the football source that are recorded the wrong way round: both legs of the tie logged at one ground, so one of them has the result reversed, across 4 seasons between 1981-82 and 2004-05. They are listed and left alone. Guessing the correction from the symptom is how a half-right fix gets shipped, so they wait for the real results. Six points deductions are modelled as deductions and not as errors, and the boards quote what a club earned on the pitch, with the deduction named beside it.
The five continental ledgers
Spain, Italy, Germany, France and the Netherlands are scored by the SAME model as England, imported rather than copied: the build asserts it is running the English ledger’s own functions, so the two cannot quietly drift apart. Results are compiled by James Curley, github.com/jalapic/engsoccerdata, with France 1994-95 supplied separately because the source does not carry it. The settings were fitted on England and reused unchanged, which is a stated choice and not a derived one.
5 seasons are recorded here whose TABLE is not final, although the ratings from them are sound: Italy 1945-46, two regional groups (Alta Italia, Centro-Sud) plus a final round; Germany 1991-92, source truncated on 1992-04-25, last 4 matchdays absent, 340 of 380; France 1946-47, source 4 matchdays short (20 clubs at 34 of 38) and one round mis-dated; France 2019-20, abandoned at the COVID-19 shutdown, 279 of 380 played; Netherlands 2019-20, voided at the COVID-19 shutdown, 232 of 306 played. They are labelled wherever their numbers appear rather than quietly averaged in.
Two units, never added
A football win was worth two league points until 1981-82 in England, and the same switch landed in 1994-95 in France and Italy and 1995-96 in Spain, Germany and the Netherlands. So league points cannot be summed down the length of one series, nor ranked across six of them. Club and metro totals use match points instead: a win is 1, a draw 0.5, which every era and every league shares. The NFL is in wins. A page that added them together would be inventing a quantity.
Sources
English top-flight results from this site’s own match log, extended from 2023-24 with football-data.co.uk, which also supplies the prices: closing where it publishes one, pre-match before 2012-13, never the two conflated. NFL results and pre-game probabilities from the site’s NFL workbook, with closing spreads loaded from published historical odds. Football ledger built 2026-09-02; NFL ledger built 2026-08-20. Player valuations from Transfermarkt via github.com/dcaribou/transfermarkt-datasets (CC0), 2012-07 through 2026-06; paused upstream since July 2026, so the series ends where the source ends.
The same measure appears one sentence at a time on every covered club page and on the metro pages, where it belongs to the club or the place rather than to the model. See it on Leicester City, Girona, Liverpool or Green Bay. Season-by-season NFL game logs live at /teams/nfl/expectation/[season].