Methodology

Every night RealStars finds candidate repos with the GitHub search API (recently created or pushed repos with many stars) plus the public GitHub Trending page. For each repo we pull privacy-safe /stargazers/history (daily counts, no identities) and sample recent WatchEvents from the public events feed for account-age signals. No repositories are cloned. (GitHub restricted the old stargazer list endpoint to admins/collaborators in July 2026.)

Signals

• Burst share: fraction of historical stars that landed on “burst days” (≥ max(200, 5× recent median daily) or ≥10k/day farm spike).
• New-account share: of recent WatchEvent actors, fraction whose GitHub user id indicates a very recently created account (id > 206000000, ~18 months).
• Low-activity profiles (when API budget allows): of up to 25 recent WatchEvent actors, those with no public repos and ≤1 follower, or an empty profile.

These follow signals described in “4.5 Million (Suspected) Fake Stars in GitHub” (ICSE 2026).

Scores

Unusual-pattern % = max(burst share, weighted account signals). Real growth = sampled stars/day × (1 − unusual-pattern %). Labels: Looks organic (<20%), Some unusual activity (20–40%), Unusual star pattern (≥40%).

Disclaimer

RealStars reports statistical patterns in a small sample of public data. A high percentage is not a determination that any person or project bought, faked or manipulated stars; launches, viral posts, events and new-user waves produce similar patterns. Figures may be stale or wrong. Maintainers can request a correction and we will review promptly.