About

The problem with every other board

Product leaderboards fail in one of two ways. Either rank is for sale — pay more, place higher — or rank goes to whoever can mobilise the most upvotes on launch day. Both measure the same thing: how much attention a founder can buy or borrow. Neither measures whether the product is any good.

An upvote is a vote about one product in isolation, which makes it almost free to give and almost meaningless to receive. There is no cost to clicking, no comparison being made, and no way to tell an enthusiastic user from a coordinated one.

Ask a harder question

So we do not ask “do you like this?” We ask which of these two is better? A forced choice between two real products is a question you cannot answer without actually having an opinion, and it produces the one kind of data a rating system can use: a result.

From there it is chess. Every matchup is a game, every vote is a result, and ratings settle where the results say they should. Beat something rated far above you and you climb a long way; beat something far below you and you barely move. Rank is won, never bought.

Two ratings, doing different jobs

Live Elo moves the moment you vote. It is what produces the +12 / −12 in the arena, and it is deliberately order-sensitive and a little twitchy, because responsiveness is the entire point of it.

Bradley-Terry owns the board. It is refit over every vote in the season at once, so the result does not depend on the order votes arrived in — the reason LMArena moved off Elo for ranking. It also means a burst of rigged votes is diluted by the whole history at the next refit rather than being permanently baked into a sequential rating.

What we deliberately do not do

• No paid placement. There is nothing to buy. That is the product.
• No ratings in the arena. The two cards show no rank and no score. Showing either would anchor you to whichever product is already ahead and poison the ratings the board is built on.
• No ranking on a handful of votes. Below 20 matches a rating is mostly noise, so new arrivals sit in Rising until the sample means something.
• No judging your own product. If you own a listing, it is never shown to you as a matchup.