Show HN: A rated ladder for vibe coders – we rank the human, not the model
21 days ago
- A weekly vibe-coding competition runs every Saturday: a secret build task drops at 17:00 CET and participants have 90 minutes to complete it using any AI tool.
- The deliverable is a single self-contained HTML file that works offline with no network calls, external scripts, or servers; participants submit the file plus their full AI session transcript.
- Entries pass a functional smoke test, then are compared pairwise in Swiss-style judging using an LLM judge against a rubric, with human spot-checks and manual audits of top finishers.
- Results feed an Elo-style/Bradley-Terry rating system; Monday brings the leaderboard, full gallery, and the winner's annotated replay.
- Season 1 deliberately offers no cash prizes—the rating, badge, and published replay are the reward, with sponsorship possibly starting from Round 2.
- Entries require a verified Google or GitHub sign-in, a public nickname, and one entry per person; no anonymous or throwaway accounts are allowed.
- Transcripts are mandatory and audited, serving both as anti-cheat proof of a human driving the session and as the competition's core content—winning replays get published and annotated.
- Judge-gaming is strictly forbidden: any visible or hidden text aimed at a judge or AI causes disqualification for that round.
- Participants retain copyright to their work, but entering grants permission to publish submissions and transcripts in the round gallery.
- The competition is designed for small, complete apps that can be built in about an hour, with a public functional checklist released after each round to show exactly what was graded.