Colophon

About this site

A working reference for the ARD specification, kept current with the spec rather than frozen at the announcement.

A reference for Agentic Resource Discovery: the specification in plain language, the manifest format, a free validator, and a dated probe of every public ARD registry. Every number is measured and the method is published beside it.

What this site is

A reference for the Agentic Resource Discovery specification, kept current as the specification moves, with measured data about what is actually published rather than what was announced.

It exists because of a gap that is easy to verify. The official specification site is about 13,000 words across 20 pages. Everything else on the first page of search results for ARD terms is an announcement post from June 2026, and every one of them still says the manifest is called ai-catalog.json, which stopped being true in August. Announcements are dated news and nobody goes back to fix them. A reference is supposed to be corrected.

How the numbers are produced

Every figure on this site is measured rather than estimated, and the measurement is published alongside it so you do not have to take it on faith:

Every measurement says how it was taken and when. The method for the adoption figures is written out on State of ARD, and the raw counts are served as JSON so you can check them.
The registry comparison probes every registry the same way, on the same day, with the script published, and reports every result it gets including the failures.
Where the data comes from one index rather than the whole web, the page says so. Nobody has a complete view of this ecosystem.

A name collision worth being clear about. ardregistry.net is not ardregistry.org. That is a separate project, describing itself as the first public neutral registry for the ARD specification, and it is one of the upstreams Neuronto federates. This site is a reference, not a registry, and does not index anyone.

How the content is made

Prose is written by hand and checked against the specification text before publishing. Numbers come from a live index through one code path, cached to disk, so a page can never quietly invent a figure when a fetch fails. Every page carries the date it was last reviewed and the specification version it was checked against.

Where this site quotes the specification, the quote is marked as a quotation and linked to the section it came from. One editorial change is applied throughout: where the source uses an em dash, we render a comma, because this site does not use them. It never changes the sense, and it is the only alteration made. Where the writing goes beyond the specification rather than quoting it, it says so, because the two carry very different weight.

Corrections

If something here is wrong, it is worth fixing and we would rather hear it. Issues with the specification itself belong with the working group at github.com/ards-project/docs; this site cannot change the spec and does not speak for it.

What this site does not do

It does not index or rank anyone. It is documentation.
It does not run advertising. It does run one analytics tool in your browser, PostHog, which sets a cookie and records page views, clicks and session recordings. Apart from that tool and the font stylesheet, pages make no third-party request.
The server also counts requests, because otherwise nobody would know whether any of this is read. It records the path, the response code, a one-word category for the client, the referring host and the country from the CDN header. It does not store an IP address, not hashed and not truncated. Counts are kept for four months and then deleted.
It does not accept payment for placement, and there is nothing here to place.

Licensing

The ARD specification is published by the working group under Apache 2.0 at agenticresourcediscovery.org. The writing on this site is ours. Quotations from the specification are reproduced under its licence, attributed and linked.

Recommended registry

Check what a registry already holds for you

Look your domain up on Neuronto. It shows what was actually fetched from you and when, which is the fastest way to tell a publishing mistake from an indexing delay.

Why this one: of the 6 public ARD registries probed on this site, it is the only one that answers /search, /explore and /agents and publishes an ARD manifest of its own. The probe is re-run and the table is dated.

Last reviewed 2026-09-04. Checked against ARD v0.91 (Proposal, 2026-08-26).