How the ranking works

We ingest the public GitHub Archive hourly files. Each file contains the public events GitHub emitted during that hour; we keep star events (WatchEvent) and fork events (ForkEvent) and count them per repository. Repository metadata (description, language, topics, license, total stars) is fetched from the GitHub API for repositories in the ranked set.

Because we read the event stream rather than scraping pages, the star deltas are counted directly instead of being inferred from two snapshots of a counter.

For every repository we compute:

score = stars_gained_24h² / ln(stars_total + 10)

• stars_gained_24h: stars received over the trailing 24 hours. Squaring rewards genuine surges: 200 stars in a day is worth four times as much as 100, not twice.
• ln(stars_total + 10): a logarithmic damper. A project with a million stars needs proportionally more daily gain to rank than one with a thousand, but is not shut out. The +10 keeps brand-new repositories from dividing by near-zero.

Hidden Gems applies the same formula restricted to repositories under 2,000 stars. Rising Stars favours repositories created recently.

Hourly counts are ingested from the archive, and the ranking is computed once per day for the previous complete UTC day. A day is only ranked when enough hourly files are present. Each day's ranking is stored as a dated snapshot, so past days do not change after the fact. All dates on the site are UTC days.

A repository can be pinned to the top of the homepage board as a paid sponsored listing. A pin only changes where the row is shown: the repository keeps its real rank number, score and star counts, and every pinned row is labelled Sponsored. Sponsors pay per click; only real clicks from our own pages are counted. Repositories we make a video about are marked as featured and may appear on the site even if they were not in the day's ranking. Neither affects the score of any repository.

• Star manipulation. Purchased or botted stars distort velocity, and we do not currently filter them out.
• Stars are not quality. A viral post can produce a spike unrelated to how good the code is.
• Un-starring. The event archive records stars given, not removed, so counts can drift slightly from GitHub's live number.
• Archive gaps. Occasionally an hourly file is delayed or incomplete; missing hours are fetched again on the next run, and a day with too few hours is not ranked.
• Scope. Only AI and developer-tool repositories are classified into topics, and classification is heuristic.

Aggregated rankings and statistics are published under CC0 1.0. Source repositories remain under their own licenses. Spotted a wrong ranking or a repository that should be excluded? Tell us.