In the ELo formula dividing by 400 means a 400 point rating advantage = A 90% chance of winning, but this number was just a guess, since large data bases didn’t exist in the 1960s when the formula was created. In 2011, using a million games, Jeff Sonas determined the number should be 480: https://en.chessbase.com/post/the-elo-rating-system-correcting-the-expectancy-tables. With a 24 k-factor, you get near perfect predictions, with a tiny formula https://en.chessbase.com/post/the-sonas-rating-formula-better-than-elo.

To this day, people continue using 400 out of tradition. Making it the base of their more complicated rating systems, like glicko (used by ogs and chess.com) and Whole History Rating (goratings.org), thinking they “updated” elo, by factoring in more values, without knowing their base is out dated. WHR also assumes 20 is the optimal k-factor (it isn’t), and then says other systems are better than standard elo. https://inria.hal.science/inria-00323349/document. They would be, if they divided 480.

Will Sonas’ discovery be lost by time?