GitHub Star Benchmarks: a dataset of 32.6 million repositories and a look at star inflation

⚡ The gist in 5 seconds - The repoboost-hq/github-star-benchmarks repository has published a dataset covering 32.6 million starred GitHub repositories. - The data is split by thresholds of 100 / 1,000 / 10,000 / 100,000 stars, as well as by language and age cohorts. - A separate section covers the research context of star buying and fake stars; everything is available in an open GitHub repository. ### 🔍 What was found A new repository, repoboost-hq/github-star-benchmarks, has appeared on GitHub — a dataset on GitHub stars covering 32.6 million starred repositories. The dataset is structured by popularity thresholds: 100, 1,000, 10,000, and 100,000 stars, providing ready-made slices for comparing projects of different scales. Additionally, cohorts by programming language and repository age (i.e., how long they have existed) are announced. The stated purpose of the dataset is to serve as a benchmark and a foundation for AI research around GitHub stars. The description specifically mentions the research context of star buying (star-buying) and fake stars (fake-star) — meaning the material is positioned not only as popularity statistics but also as a basis for analyzing star inflation. No specific metrics on the share of inflated stars, counting methodology, or snapshot date are provided in the original description — only the stated dataset structure and a link to the repository are available. ### 💡 Why it matters GitHub stars have long become a de facto trust metric: people look at them when choosing libraries, investors and recruiters take them into account, and search rankings rely on them. The threshold slices of 100/1k/10k/100k turn a raw list into a tool for testing hypotheses — for example, about how repository age correlates with star accumulation or how dynamics differ across languages. The separate layer on star buying matters for those building project scoring systems who want to distinguish organic growth from inflated numbers. That said, it is worth noting: the repository is published under the name repoboost-hq, which hints at the repo promotion space, so it makes sense to treat the conclusions about fake stars with the authors' potential commercial interest in mind and to verify the methodology against the raw data.

🔗 Read on github.com