# CrawlCheck trust profile for ARD entries, v1

Status: in use on crawlcheck.io. Updated 2026-11-13.

Agentic Resource Discovery (ARD, v0.91) lets an entry carry a `trustManifest`. ARD requires one member, `identity`, and leaves signing and verification to the framework the manifest names in `trustSchema` (ARD sections 4.5.1 and 4.5.2). This page is that framework for entries that name CrawlCheck: what the attestation says, how to check it, and what it does not say. ARD is an independent specification; CrawlCheck did not write it and this profile is not part of it.

## 1. The block

```json
"trustManifest": {
  "identity": "https://example.com",
  "identityType": "https",
  "trustSchema": {
    "identifier": "crawlcheck-data-v1",
    "governanceUri": "https://crawlcheck.io/spec/ard-trust",
    "verificationMethods": ["Ed25519"]
  },
  "attestations": [
    { "type": "crawlcheck-site-verification",
      "uri": "https://crawlcheck.io/api/v1/resolve?domain=example.com",
      "mediaType": "application/json" }
  ]
}
```

`identity` is the publisher's own domain, the same domain as the `<publisher>` in the entry's `urn:air:` identifier. The live example is every entry of https://crawlcheck.io/.well-known/ard.json.

## 2. What the attestation is

The `uri` returns CrawlCheck's signed Resolve answer about that domain: what CrawlCheck measured there, when, and the decisions that follow for read, cite, connect and transact. The full document is specified at /spec/resolve.

It states three things a registry can use:

- **Who issued it.** The answer is signed by a key published by crawlcheck.io.
- **That it is intact.** The signature covers a hash of every field.
- **Whether control of the domain was proven.** The `control` section reads `proven` when the domain's holder showed control to CrawlCheck (a DNS TXT record, a homepage meta tag, a well-known file or a Cloudflare zone), with the method and the date.

## 3. How to verify one

1. Fetch the `uri`. Check `domain` equals the publisher domain of the entry.
2. Remove `sha256` and `signature`, serialise the rest as canonical JSON (keys sorted, no whitespace) and take its SHA-256. It must equal `sha256`.
3. The signed message is the text `crawlcheck-data-v1`, a line feed, then that hash in hex. Verify `signature.sig` over it with Ed25519.
4. Take the key by `signature.kid` from https://crawlcheck.io/.well-known/http-message-signatures-directory. Do not trust the copy of the key inside the answer on its own.
5. Read `answered_at` and `freshness`, and decide how old an answer you accept.

The reference verifier does all of it with no dependencies:

```
curl -s 'https://crawlcheck.io/api/v1/resolve?domain=example.com' > answer.json
curl -sO https://crawlcheck.io/crawlcheck-verify.mjs
node crawlcheck-verify.mjs --online answer.json
```

Exit code 0 and the word ACCEPTED mean the answer is intact and signed by a published CrawlCheck key.

## 4. Publisher binding (ARD 4.5.1)

ARD asks that the trust domain in `identity` align with the publisher in the URN. This profile keeps two cases apart:

- **Entries published by crawlcheck.io.** Publisher, identity and issuer are the same domain. The attestation is issued by the publisher.
- **Entries published by another domain.** The attestation is a third party's signed measurement of that domain. It supports the binding only when `control.state` is `proven`; without that it says what CrawlCheck measured and nothing about who controls the domain.

## 5. What it does not say

- It is not a rating, a ranking or an endorsement of the resource. ARD's relevance score and this attestation are separate things.
- ACCEPTED is about who signed the answer. What an agent may do is in the answer's `actions`, and a decision for one action comes from `POST /api/v1/preflight`.
- An answer describes the domain at `freshness.observed_at`. It does not say the site is unchanged since.
- It does not attest the artifact behind the entry (a server card, a skill file). It attests the domain that publishes it.

## 6. Adding it to your own entries

Any publisher may add the block in section 1 with its own domain in `identity` and in the `uri`. The answer is public and needs no key. Its `status` says how much was measured: `measured` after a full scan, `partial` before one, so scan the domain first at https://crawlcheck.io. To make the answer support the binding, prove control of the domain (section 2). For a domain CrawlCheck has scanned, `GET /api/fix/ard?domain=example.com` drafts a manifest with the block filled in.

## 7. Changes

- 1 (2026-11-13): first version.
