Who claims to be an agent, who vouched for them, and how much of either turns into anything measurable. A registration is a claim, not activity.Note 23
ERC-8004 identity registry · the 204,760 registrations whose metadata document is held INLINE on chain · pointer cards are not followed
An agent card is the one place a registration says anything about itself. Most of what it says cannot be checked from the registry, but WHETHER it says anything can, and mostly it does not.
Declared against observed. 5,231 registrations name another registry they are in. We observe 30,440 held by an address that also holds identities on another chain, and 144,355 describing themselves with a document used on another network. Agents are far more cross-chain than they say they are, which is what makes the declared field close to useless for finding them and the observed link worth computing.
Only inline cards are read. 1,373,527 registrations point at a document somewhere else and we do not follow the pointer: a count that depends on somebody else’s server answering is a count that changes when their server is down, and we would have no way to tell that apart from an agent changing its mind.
A host with hundreds of registrations invites the question a count of hosts cannot answer: hundreds of agents that happen to use the same provider, or one deployment wearing hundreds of identities? The registry cannot tell them apart. The addresses holding the tokens can.
| Host | Registrations here | Held by | Each holding | Reads as |
|---|---|---|---|---|
| github.com | 9,968 | 944 | 10.6 | a handful of parties |
| api.celonova.xyz | 8,490 | 5 | 1698.0 | a handful of parties |
The 20 largest of 2,611 distinct hosts, across 15,662 registrations. An endpoint is a claim like any other, and now a checked one.
1,594 of 2,328 hosts asked no longer resolve at all. 1,901 agents point at a domain that is gone.
A timeout describes the attempt, not the host — a slow endpoint, our network, or a firewall that drops rather than refuses all look the same from here, so they are counted apart rather than called failures. 2,328 of 2,610 hosts were reached this run; the rest were not asked rather than found absent.
5 of 720 answering hosts serve a card naming an address that registered there. Everything else on this panel is a claim the domain has not corroborated.
Asked at the two paths the A2A convention uses, and only of hosts that answer. A domain that does not resolve has nothing to say about control, and the panel above already reports it. An address is counted only where the served card puts it in a field naming a party: a payment token is not an identity, and reading one as a contradiction was this check’s first mistake.
Counted 2026-10-25 10:15 UTC, from 57,751 distinct documents; 204,760 inline registrations share them, so a claim is counted once per document and once per registration behind it, never weighted by how often it is repeated.
| 8,485 |
| 2 |
| 4242.5 |
| a handful of parties |
| chainpulse-mcp-production.up.railway.app | 477 | 130 | 3.7 | a service several parties chose |
| ensoul.ac | 379 | 34 | 11.1 | a handful of parties |
| Analyst | 341 | 5 | 68.2 | a handful of parties |
| Analyst | 320 | 282 | 1.1 | a service several parties chose |
| Analyst | 315 | 2 | 157.5 | a handful of parties |
| Analyst | 258 | 256 | 1.0 | a service several parties chose |
| Analyst | 231 | 231 | 1.0 | a service several parties chose |
| Analyst | 230 | 230 | 1.0 | a service several parties chose |
| Analyst | 105 | 1 | 105.0 | one party, many identities |
| Analyst | 102 | 87 | 1.2 | a service several parties chose |
| Analyst | 101 | 101 | 1.0 | a service several parties chose |
| Analyst | 87 | 76 | 1.1 | a service several parties chose |
| Analyst | 87 | 69 | 1.3 | a service several parties chose |
| Analyst | 70 | 1 | 70.0 | one party, many identities |
| Analyst | 58 | 57 | 1.0 | a service several parties chose |
| Analyst | 57 | 2 | 28.5 | a handful of parties |
| Analyst | 55 | 55 | 1.0 | a service several parties chose |
2,442 of 2,611 hosts are named by registrations a single address holds, covering 3,005 of 15,378. None of this is an accusation: running many agents from one deployment is a reasonable thing to do, and the point is only that the registry counts them as many and nothing in it says so.
And nothing on any card can be checked cryptographically. 18 of the cards read declare a trust model they support. 1 publish anything under one that a reader could verify: a signature, a JWS, an attestation. Saying which model you support is free; publishing something somebody can test is the part nobody has done. A field naming a signer is not a signature, and is not counted here as one.
| Host | Registrations naming it |
|---|---|
| github.com | 9,968 |
| api.celonova.xyz | 8,490 |
| celonova.xyz | 8,485 |
| chainpulse-mcp-production.up.railway.app | 477 |
| ensoul.ac | 379 |
| Analyst | 341 |
| Analyst | 320 |
| Analyst | 315 |
| Analyst | 258 |
| Analyst | 231 |
| Analyst | 230 |
| Analyst | 105 |
| Analyst | 102 |
| Analyst | 101 |
| Analyst | 87 |
| Analyst | 87 |
| Analyst | 70 |
| Analyst | 58 |
| Analyst | 57 |
| Analyst | 55 |
5 of 20 shown in full. 15 more are listed with the hosts themselves withheld. The count is free because the count is the finding, and so is every reading on any vault’s own page. What Analyst carries is the SCREEN: all 20 of them in one table, filtered and sorted and compared, instead of one vault at a time.
25 distinct registries named, by CAIP chain. A registration naming a chain is not evidence it is registered there; it is evidence it says so.
| Named registry | Registrations naming it |
|---|---|
| Celoeip155:42220 | 3,568 |
| Baseeip155:8453 | 808 |
| BSCeip155:56 | 289 |
| Ethereumeip155:1 | 247 |
| 184 | |
| Avalancheeip155:43114 | 18 |
| Monadeip155:143 | 17 |
| 14 | |
| Arbitrumeip155:42161 | 14 |
| Polygoneip155:137 | 11 |
| Abstracteip155:2741 | 8 |
| 8 | |
| Mantleeip155:5000 | 5 |
| Robinhood Chaineip155:4663 | 4 |
| Tempoeip155:4217 | 3 |
| solana:5eykt4UsFv8P8NJdTREpY1vzqKqZKvdp unrecognised | 3 |
| eip155:10143 unrecognised | 3 |
| Gnosiseip155:100 | 2 |
| 2 | |
| Taikoeip155:167000 | 1 |
| Optimismeip155:10 | 1 |
| Lineaeip155:59144 | 1 |
| Scrolleip155:534352 | 1 |
| eip155:4326 unrecognised | 1 |
| 1 |