didrod205/fake-star-detector

fake-star-detector ⭐

Is this GitHub repo's hype real? Crawls a repo's recent stargazers and flags purchased/bot stars — from the timing spikes (hundreds in an hour) and the empty accounts behind them (days old, zero followers, zero repos, starred one thing and vanished). A fake-o-meter 0–100, with the receipts. No API key needed; a token just raises the rate limit.

🌐 Check any repo + today's trending leaderboard →

Paste a repo and it analyzes the stars live in your browser. Same engine in your terminal:

npx fake-star-detector facebook/react
npx fake-star-detector trending     # rank today's trending by how bought they look

  some-trending-org/rocket-launch · 48,201 stars

  82/100 ████████████████░░░░ Reeks of bought stars 🤖

  timing (of 1,000 recent stars)
    biggest burst: 412 stars in one hour · 2026-06-12T03:00Z
    91% of them landed on a single day  ▁▁▁▁▁█▁▁

  accounts (sampled 30 recent stargazers)
    26/30 look bot-shaped (87%)
      @xk29fja — account 2d old when it starred, 0 followers, 0 public repos, no bio
      @qz81mn0 — account 1d old when it starred, 0 followers, 0 public repos, no bio

heuristic, not proof — fast bursts + empty accounts ≈ purchased stars. public data only.

## Why

GitHub stars are a currency — and currencies get counterfeited. You can buy a
thousand stars for pocket change, and a lot of repos do, especially the ones
racing up *trending*. The fakes leave two fingerprints this tool reads:

1. **Timing.** Real stars trickle in; bought stars arrive in a wall — hundreds in
   an hour, the whole batch in a day.
2. **The accounts.** A real stargazer has a history. A bought one is a shell:
   created days ago, zero followers, zero repos, one star, gone.

measures both, and gives you a score and the receipts.

How it works

GitHub stargazers API (starred_at)  ─crawl recent→  timing spikes  ─┐
GitHub users API (account metadata) ─sample→  empty-account share  ─┴→  fake-o-meter

• Browser-safe — api.github.com allows CORS, so the exact same crawler runs

in the CLI, in CI, and live in the web page (no node: in the core).

• Account signal dominates the score (empty accounts are the real tell); the

timing concentration and single-hour burst corroborate. Among trending repos a big burst is expected, so we don't cry wolf on timing alone.

• The live leaderboard is static: a GitHub Action re-analyzes trending every

few hours and commits the result — no server.

Install & usage