Back to voting

How the table works

Every vote is a match

There's no like button on GOAT. You get two names and pick one, and the table is worked out the way chess ranks its players: each pick is a result between two rivals, and a win counts for more the stronger the opponent.

Why not upvotes and downvotes?

Upvotes answer the wrong question. "Do you like Pelé?" gets a yes from almost everyone, and so does "Do you like Messi?" A list built from those answers ranks who is famous and who gets seen most, not who is greatest. The argument we're here to settle is Pelé or Messi, and only a head-to-head asks that.

Picking between two is also easier and more honest. Nobody can say how many points Federer deserves out of ten, but almost everyone has an instant answer to Federer or Nadal. Chess found the same thing seventy years ago: you can't score a player in a vacuum, but you can see who they beat.

A tally of votes can't tell a hard win from an easy one. Beating the number one should matter more than beating the number ten, and losing to the number one should cost almost nothing. A rating does that on its own; a vote count never will.

And totals reward whoever has been around longest. A contender added last week starts thousands of votes behind and stays there. A rating has no head start to protect: a newcomer who keeps winning climbs in days.

Side by side

Upvote / downvoteHead-to-head rating
MeasuresHow liked or how visibleWho beats whom
OpponentIgnoredA win over a strong rival is worth more
NewcomersStart far behind, stay behindMove fast until we know where they belong
Pile-onsDownvote a rival straight to the bottomNo down button: you can only pick between the pair you're shown
CertaintyHiddenShown as ± next to every rating

How a vote moves the table

We use Glicko, Mark Glickman's 1995 refinement of the Elo system chess federations have used since the 1960s. Chess.com and Lichess run versions of it today.

  1. Everyone starts at 1500 ± 350

    1500 is the middle of the scale. The ± is how unsure we are: at 350, the real rating could be almost anywhere.

  2. Each matchup has an expected result

    The rating gap predicts the winner. 200 points apart, the favourite is expected to win about 3 times in 4. Both sides gain or lose by how far the real result beats or misses that prediction.

  3. Surprises move it most

    A favourite who wins as expected barely moves. An upset moves both sides a lot. So does anything involving a contender we're still unsure about.

  4. The ± shrinks as votes come in

    Every result makes us surer, so the ± narrows and each later vote moves that contender a little less. It never drops below 30, so the table can always change its mind.

Worked examples: one matchup

MatchupExpected to winResultWinnerLoser
1900 ± 50 vs 1700 ± 5076% · 24%Favourite wins+3−3
1900 ± 50 vs 1700 ± 5076% · 24%Upset+11−11
1800 ± 50 vs 1800 ± 5050% · 50%Either wins+7−7
Newcomer 1500 ± 350 vs 1900 ± 509% · 82%Newcomer wins+473−8

The last row is the point of the ±. One shock win sends an unproven newcomer flying, because we knew almost nothing about them. The established favourite barely drops, because losing to an unknown says little about them. The two "expected" figures don't add to 100% there: each side's prediction is discounted by how unsure we are about the other.

Try it

Set two contenders and pick a winner. These are the same sums the table runs.

Contender A
Contender B
Winner

What keeps it fair

  • Recalculated every five minutes. Votes are collected, then the whole table is rated in one go. Each pair counts once per round, scored by its share of wins, so a matchup that gets hammered with votes doesn't count extra.
  • You don't choose the matchups. A matchmaker deals each visitor their own stream: least-seen and least-certain contenders first, close contests preferred, one wildcard in seven, never the same pair twice. You can't go looking for a rival to sink.
  • Limits on hammering. 40 votes a minute per voter and 120 per network address, and repeating the same pair within ten seconds isn't counted.
  • Money never buys rank. Sponsors can put their name on a bracket. Results come from votes only.

The "Week" column on each table compares today's positions with a snapshot taken a week ago. Vote counts, ratings and generation splits on the prototype are mock data until the site opens.

Go and vote