GitHub's fake star economy
GitHub’s star counts, once treated as a rough proxy for an open‑source project’s popularity and quality, are increasingly being gamed through paid “fake stars” and hackathon-style incentives. Commenters argue this has corrupted a signal that some investors, recruiters, and developers still use to gauge traction, illustrating Goodhart’s law: once a metric becomes a target, it ceases to be useful. Many suggest looking instead at harder‑to‑fake indicators such as active contributors, issue and PR history, dependency usage, and real‑world adoption, while noting that any widely used metric is eventually vulnerable to manipulation.
Perceived Problems with GitHub Stars
- Many commenters see stars as a very weak signal: costless to give, easily faked, and not tied to real usage or quality.
- Goodhart’s law is cited repeatedly: once stars became a target for VCs, employers, and marketing, they stopped being a good measure.
- Several describe firsthand evidence of repos with huge star counts but almost no issues, PRs, forks, or meaningful commits.
- Some now treat a very high star count (especially for AI/agent projects) as a negative signal for hype.
Why VCs and Others Still Use Them
- Stars are simple, numeric, and legible to non-technical investors and committees; they help justify decisions.
- For early-stage OSS startups, there often aren’t better easy numbers; stars, downloads, and social buzz become proxies for “traction.”
- Some argue sophisticated funds mostly discount stars now and do deep diligence; others say the broader ecosystem and tooling (indexes, scrapers) still over-index on them.
Alternative Signals and Heuristics
- Common replacement heuristics:
- Recent commit activity, project age, and commit history.
- Issue volume, quality, and maintainer responsiveness.
- Number and identity of contributors; “bus factor.”
- Release cadence, changelogs, dependency hygiene, and API elegance.
- Forks and who stars/forks the repo, not just how many.
- Several suggest graph-based or reputation-weighted metrics (PageRank/“peoplerank”-style, trusted contributor sets, network centrality).
- Others emphasize the only truly reliable metric: “does it solve my problem, and are maintainers responsive?”
Gaming, Detection, and Countermeasures
- Star-buying markets, hackathons that require starring, and astroturf campaigns are reported.
- Some propose fork-to-star ratios and zero-follower/zero-repo stargazer rates as heuristics for fake stars; others argue these signals are noisy or flawed.
- With LLMs, commenters expect next-round attacks: fake issues, PRs, and “activity” will be easy to mass-generate.
- Several believe GitHub could crack down using internal signals but has little incentive, given its social-network-like incentives.
Broader Reflections
- Many note that every popularity metric (downloads, followers, reviews, traffic) is now routinely gamed; an entire industry sells fake “signal.”
- Some still defend stars as “better than nothing” for rough discovery, especially at extremes (0 vs thousands).
- Others are moving to treating stars purely as personal bookmarks and ignoring counts entirely.