Arpad Elo rated the surprise, not the result, and chess got one comparable number

Before 1960, American chess players were rated by Kenneth Harkness's system, which averaged tournament performances but could not meaningfully compare players who never met the same opposition, and it produced ratings widely considered inaccurate. Arpad Elo, a Marquette University physics professor and chess master, studied rating systems for the United States Chess Federation and wrote up the replacement — 'The USCF Rating System' — in the March 1960 issue of Chess Life.

The design choice was to rate the surprise rather than the result. The difference between two ratings maps to an expected score, and after each game the ratings move in proportion to how far the actual result departed from that expectation. The system needed no central authority to re-center the scale, because every game transfers rating points from one player to the other, and lookup tables made it computable by hand by club volunteers.

The USCF implemented it in 1960 and the World Chess Federation (FIDE) adopted it in 1970. The 400-point scale — a 100-point gap implies roughly a 64 percent expected score, 200 points about 76 percent — became the common language of head-to-head skill, and the expected-score idea spread to football, basketball, baseball, other board games and esports matchmaking.

Update ratings by the size of the surprise

The gap between two numbers predicts the result before a piece moves

Updates are zero-sum, so the scale never needs central re-centering

Only performance against expectation counts, so results aggregate across clubs and years

Lookup tables made it runnable by volunteers decades before computers

Score what you didn't expect. If every result updates a number by how surprising it was, results from different places and different times fold into one comparable scale that corrects itself.

Elo's scale became the world standard for chess and the template for matchmaking far beyond it. The assumptions were later refined rather than replaced: chess performance is not really normally distributed, and the logistic form proved easier to compute. Glickman's Glicko and Glicko-2, which add a measure of rating reliability, underpin online play, and today's US Chess procedure runs six rating systems descended from the same expected-score logic. Updating a rating by the size of the surprise is now the default wherever two sides meet head-to-head.