The expected result
Favourite1600.0+5.81605.8
Underdog1400.0−5.81394.2
The method
A rating that rewards one thing: having won.
One engine, one rule, every discipline. Not the prize money, not the tier, not the seeding: a rating moves on results.
The idea
A rating is not a total. It is a prediction.
It answers one question: if these two meet now, who wins, and how settled is it? Everything is in the gap between two ratings.
Rating gapThe higher rated wins
050%
+10064%
+20076%
+40091%
Four hundred points apart: ten to one. That is the definition of the scale.
One result
Every match is a transfer
What the winner takes, the loser gives up, to the point. Rating is only ever won off someone.
The upset
Underdog1400.0+18.21418.2
Favourite1600.0−18.21581.8
Beating the favourite moves three times as much: the model gave it one chance in four. An expected win pays almost nothing.
The settings
The four numbers the engine runs on
There are no others, and none is tuned per discipline. A rating you cannot recompute yourself is a rating to take on trust.
1,500
Where every competitor starts.
24
The most one result can move a settled rating.
48
Doubled over the first 10 matches.
400
The gap worth ten to one.
A walkover, a retirement and a disqualification move no rating: none says who plays better. The match is still recorded, at zero.
Where it comes from
Elo was not built to reward. It was built to measure.
Árpád Élő, a physicist and chess player, chaired the US federation's rating committee. The scale then in use produced disputed ratings; he built another.
- 1903
Árpád Élő is born in Hungary. Physicist at Marquette, eight times Wisconsin chess champion.
- 1960
The US chess federation adopts his method, replacing the Harkness system.
- 1970
FIDE adopts it: the scale goes international.
- 1978
He turns the scale on the past and rates champions long dead.
- Today
Go, backgammon, football, video games. It travels because it assumes almost nothing.
A measurement has no opinion about how important a match was. That is the reason to use one.
The site owes him its name.
What makes the number checkable
Replayed, never patched
The model is applied match by match in chronological order. Every move is kept with the match that caused it.
From any profile, you can walk back to the match itself.
Extended, never mutated
Each cycle resumes the last state and applies what followed. Extending gives the same ranking as a full replay.
Three dates, never conflated
When the source was last read, the date of the most recent result, and the date the ranking was settled. A source read ten minutes ago may have published nothing for three weeks.
The check
What the calculation gave, what happened
Rating baseRightUpsetsBrierMatches
Darts · Open the matchup
65%
won by the favourite
2,459 to the underdog
0.215
Brier (0.25 for a coin toss)
7,095
matches scored
4,167 not scored
CalculatedObservedmatches
53%52%1,292
58%57%1,247
62%64%1,166
67%67%959
72%67%735
77%73%612
82%82%449
87%88%361
92%91%226
96%92%48
evencertain
Explore
For each discipline: the ranking, its sources, and what is missing.