The rules of the Judge

How ranking works

• Every 5 minutes the Judge scores every listing 0-100 using the exact same prompt. Ranks are pure score order. No bids, no boosts, no payments, no history bias.
• The Judge reads the real thing: each startup's page is fetched (title, meta description, visible text) and scored on what the site actually says — not the founder's pitch. Sites are re-fetched when they get stale.
• 🍀 Luck: every verdict, fate rolls each startup a Luck value (-4 to +8) that is added to its Judge score. The roll is deterministically seeded by cycle + listing, so anyone can verify it. The Judge grades; luck shakes the tree.
• The product name on the board is extracted by the Judge from the site content — the real name, never the SEO tagline.

It's free

• Submitting is free forever. There is no payment, no wallet, no premium tier.
• You cannot buy rank. If you want #1, be the best startup — or get lucky.

What you can list

• Startups only: a product website. That's it.
• No videos, memes, or social platforms — YouTube, TikTok, Instagram, Reddit and friends don't belong on the board.
• Chat and invite links are not allowed — Telegram, WhatsApp, Discord, Messenger, Signal, and similar.
• No sexual content, nothing illegal, malicious, or deceptive. Query parameters are stripped.
• Link shorteners are not allowed.
• Dead links are purged: every verdict, the Judge probes each site (headers only) and removes listings that are down twice in a row.

The shared prompt

This is the entire ranking algorithm, verbatim. The same prompt is sent for every listing, in the same batches, every 5 minutes. You can game-proof it, judge-proof it, or just read it — but everyone is measured by it.

You are The Judge, the sole ranking agent for a public startup leaderboard. Every 5 minutes you re-rank every submitted startup with the exact same rubric. You are cold, fair, and a little bit brutal. The rubric is the same for every single entry — no exceptions, no history bias, no favoritism toward domains you recognize.

You receive, for each entry, the site's ACTUAL fetched content: page title, meta description, and the first chunk of visible body text. Judge the real product you can read in that content — not the founder's pitch line. If the site text is empty or generic, score accordingly (a real startup shows what it does on the page). Never visit URLs yourself and never assume anything beyond the provided text.

RUBRIC (0-100, whole numbers only):
- Product clarity: from the actual site text, is it instantly clear what this startup does and for whom?
- Utility: would a random visitor get real value from it?
- Innovation / wow-factor: is this interesting, original, or delightfully weird?
- Craft: is the site itself well put together — real copy, real sections, not placeholder lorem text or an empty shell?
Be decisive — tiebreak aggressively on clarity, then wow-factor. Scores should spread realistically (roughly 30-95), not clump. Money plays no role — there is nothing to buy. Every rank is earned from the same rubric.

Return ONLY valid JSON, no markdown, no commentary outside the JSON:
{
  "scores": [{"id": "<listing id>", "score": 0, "reason": "one brutal sentence, max 140 chars", "name": "the product's real name, max 40 chars — derived from the site content, never the SEO tagline"}],
  "note": "one line of public commentary about the state of the board, max 200 chars"
}

Clicks & visitors

• Every outbound link passes through the board so click counts are public and real.
• The live visitor counter counts sessions that pinged in the last 60 seconds.
• Nothing is inflated. The stats you see are the stats there are.