How the archive is built

Every product is read from one of two places, and nothing else. 55 of them come through the GitHub Releases API, using the tags their maintainers publish. The rest come from the RSS or Atom feed the vendor publishes on its own changelog page.

4 of them write no notes in the release itself. They write a sentence naming their CHANGELOG, and the notes are in it, so the collector follows that sentence and reads the section belonging to the version out of that file. The document is the publisher’s own, in the same repository the release came from, so what arrives here is still what they wrote. A release whose body already describes the change is left alone. They are Kubernetes, Visual Studio Code, Vite and Vue.js, and every other product is read from the release itself.

No page is scraped and no browser is driven. If a product stops publishing a machine readable changelog, it stops updating here, and the archive keeps what it already collected.

Markdown is turned into HTML and the HTML is stripped down to text, links, lists, tables and code. Scripts, styles and images are removed. Headings are pushed one level down so they sit under the page title. The contributor roll that GitHub appends to generated notes is dropped, because on a large repository it can be longer than the notes themselves.

Notes longer than 40 kB are cut, and those pages say so and link to the publisher. Nothing is summarised, reworded or translated. The wording on a release page is the wording its author shipped.

Labels like added or security are matched from words that appear in the notes. They are a reading aid, not a claim about the release. A release that changed an API without using the word breaking will not carry that label.

Several products worth tracking publish a changelog for people to read but no feed for machines. Stripe and the OpenAI platform docs are the ones asked for most often. They are on the list and they stay off the site until a public feed exists or the vendor offers one. Linear and the Claude Platform notes were on that list until August 2026, when a feed appeared behind both changelogs, and they are tracked here now.

A product qualifies when it has a public GitHub releases page or a public RSS or Atom changelog, and when its releases carry notes rather than bare tags. Send the feed address and the reason it belongs here to hello@traceary.com.

Release notes stay the property of the projects and companies that wrote them. Traceary stores them for reference and links back to the original on every page. Any publisher who wants an entry removed can write to the same address and it will be taken down. Copyright and removal covers who owns what, what to put in the request, and what happens to the entry afterwards.

The archive is served as data as well as pages. catalog.json lists every product with its source and its counts, each product has a releases.json holding its versions, dates and labels, and status.json reports whether collection is running. Every product also has an RSS feed, and llms.txt is a map of all of it, including the things the archive does not claim to know.

The release notes themselves are left out of the JSON. They are their publishers’ words, they are on the pages with a link to the original beside them, and handing eleven megabytes of someone else’s prose to a machine is a different act from publishing it on a page that credits them.

The status page records what the collection job managed to read, source by source, so the claims on this page can be checked rather than taken. Traceary’s own releases are on its changelog, and the full list of tracked products is in the catalog.