How it works

The leaderboard uses ELO - the same system that ranks chess players. Instead of scoring each portfolio in a vacuum, it only ever asks one question: “which of these two is better?” Every answer nudges the two sites' ratings.

Everyone starts at 1200. When you pick a winner, it gains points and the loser drops by the same amount - but how many points depends on the upset. Beat a site rated far above you and you jump a lot; beat one far below and you barely move (you were supposed to win). That self-correcting math is why a handful of votes can sort thousands of sites sensibly.

Two twists specific to here: AI seeds, humans refine. Before any people vote, Gemini casts the first rounds so the board isn't empty on day one; then real votes take over. Superstars let you spend an earned ⭐ to make a vote count double - for the rare site you think is genuinely exceptional.

And because every vote is logged, the entire ranking is recomputable - if a bad actor is ever found, their votes can be removed and the board rebuilt from scratch.

Ranking 1,700 sites by hand is a non-starter, so an AI does the first pass. Each site gets a screenshot captured by a real browser; Gemini then looks at it and scores six things - visual design, the 5-second test (is it clear who you are?), project storytelling, writing, memorability, and motion.

For the leaderboard, the AI also votes head-to-head, the same way you do - but it weighs more than looks. Its comparison folds in the objective scorecard (Lighthouse performance + accessibility, plus polish signals like a custom domain and a working share card). A gorgeous page that's slow or inaccessible loses ground it would've won on looks alone.

The honest caveat: an AI judging a screenshot judges how something looks - not whether the projects are real or the code is good. That's exactly why it only seeds the ranking and humans get the final say. The model and prompt are fixed (one model, for consistency) and every AI vote is tagged, so its influence can always be separated out.