How we research

Four parts: what collects the data, what calculates from it, what an LLM does, and what a person does. All times are UTC.

Automated collection

Scheduled jobs read public sources and store counts, links, titles and dates. They do not store the text of posts or articles.

• GitHub: hourly, at 5 minutes past. Stars, forks, issues, watchers and activity. Once a day, contributors and the latest release.
• GitHub health: daily at 04:30. Commit activity, issues, releases and community files.
• Hacker News: every 15 minutes.
• Hugging Face: hourly, at 15 minutes past. Likes and downloads for models and spaces.
• Packages (npm, PyPI, Docker Hub, crates.io, Homebrew): daily at 03:30. Downloads, pulls and installs, linked to the GitHub project.
• deps.dev: daily at 05:00. OpenSSF Scorecard, dependents and advisories.
• Newsletter and vendor feeds (RSS): every 2 hours, at 20 minutes past. We record only that an item links to a project we track.
• Discovery feeds: every 30 minutes. New projects from GitHub search, Hacker News, Hugging Face, OpenRouter, the MCP registry, npm and Lobsters.

Deterministic calculations

Velocity, baselines and labels are fixed rules applied to the stored numbers. The same data gives the same label every time. The rules are written out on the methodology page.

LLM-assisted tasks

None today. No language model writes, scores or labels anything on this site. If that changes, this page will say where.

Human review

An editor can merge duplicate projects, and hide or remove a project. Merges are recorded with the date and a reason. Removals are recorded with a reason, and the project's stored numbers are deleted. Corrections are listed on the corrections page.