Data sources
Where each number comes from, how often it is refreshed and what to keep in mind when reading it.
All sources below are public and are collected once a day. The collection code reads the product's identifier (package name, repository, extension id and so on) from our registry and stores one row per day per product.
Package registries
npm. Daily download counts from the npm downloads API, with up to 18 months of history. Counts include every download of the package, automated ones included.
PyPI. Daily download counts from the public PyPI download logs, queried through ClickPy, with pypistats.org as a fallback. The newest day is still filling up when we collect, so the series stops two days back.
Homebrew. Install counts over 30, 90 and 365 days from the Homebrew formulae API. Homebrew analytics are opt-out and cover macOS and Linux users of Homebrew only.
Docker Hub. Cumulative pull count of the official image from the Docker Hub API.
Code hosting
GitHub repositories. Stars, forks, open issues, licence and language from the GitHub REST API. Star history is built from our own daily readings.
GitHub releases. Tag, date and release notes of each release. Pre-releases, nightlies and release candidates are filtered out of the counts and feeds. The bullet points shown on the site are shortened from the project's own notes and link back to the full text.
Pull requests by coding agents. Weekly counts from GitHub issue search of public pull requests that match an agent's signature, such as its bot account or branch prefix, and of those that were merged. Private repositories are not visible.
Marketplaces
Visual Studio Marketplace. Install count, average rating and number of ratings for the product's VS Code extension.
Open VSX. Download count and rating from the Open VSX registry, which serves editors built on the open-source VS Code codebase. Its count includes update downloads, so it is reported separately from VS Code installs.
JetBrains Marketplace. Plugin download count.
WordPress.org and the App Store. Active installs and ratings of official plugins, and rating counts of official iOS apps, where a product has one.
Attention
Hacker News. The number of stories and comments per day that mention a product, counted on the public Hacker News dataset hosted by ClickHouse. Products whose names are common words are not counted.
Wikipedia. Daily views of the product's English Wikipedia article from the Wikimedia Pageviews API, excluding known bots.
Tranco. Daily rank of the product's own domain in the Tranco list, a research-oriented ranking of the most visited sites that combines several providers and is hardened against manipulation.
Reliability
Vendor status pages. Incident history from the JSON API of the vendor's public status page, where it runs on a platform that exposes one. We keep our own log, since the APIs return only recent incidents.
What we do not use
We do not scrape review sites, estimate revenue or user counts, or use sampled panel data. If a number cannot be traced to a public source, it is not on the site.
Download the data
The current snapshot is available as JSON, and each agent's page links to its daily series as CSV. If you use the data, a link back is appreciated.