Loading
0.5%
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
The standard has three registries. We read 329,320 registrations out of the first and hundreds of thousands of ratings out of the second. The third (the only one of them meant to carry evidence rather than a claim or an opinion) is deployed on 4 of the chains we checked and has emitted 2 events in 3,600,000 blocks.
A registration is a claim, a rating is somebody’s opinion, and a validation is the only one of the three that is supposed to be evidence. It is the load-bearing half of the standard and it is unused: 3,600,000 blocks were swept across 4 chains for 2 events.
Probed rather than ingested, and that is a decision. A reader for a contract with two events would be a pipeline for an empty table; the emptiness is the finding, and this platform publishes those rather than waiting for them to become interesting. If it starts filling, the sweep says so and the reader becomes worth writing.
The blocks scanned are published beside the count on purpose. “No events” over an unknown span is not a measurement.
ERC-8004 identity registry against the three rails we read · ACP on Base · the Olas mech marketplace on Gnosis · x402 settlements
698 of 329,320 addresses holding a registered identity have ever appeared on a rail that moves money; 0.2%. The registry and the working economy are very nearly disjoint populations, and every chart that draws them as one funnel implies otherwise.
This is a floor, and the reason is worth reading. On the registry side the address is the holder of the identity token; on the rail side it is whoever paid or was paid. An operator holding its identity with one key and trading with another reads as absent here. For the cards that name a wallet the two are the same address, which is why the comparison is possible at all, but nothing forces that, so the overlap can only be understated.
Three rails, not all rails, and not the same three as everywhere. This counts ACP, the Olas mech marketplace and x402 together. The figure under Registration & ratings counts ACP alone and is an order of magnitude smaller for that reason. An agent earning somewhere we do not read is invisible to both, exactly as it is to every other count on this site. It is the same distinction the declared-registry tab makes between a claim that fails and a registry we never looked at.
No value column spans the rails. ACP carries a dollar figure per job; the mech marketplace is priced in a token no feed we hold prices; x402 settles in several tokens at once. One “value” column across the three would be three units added together.
A payer and a provider are different populations and are never merged: buying something and being paid for something are different claims about being an agent.
| Rail | Doing what | Addresses | Hold an identity | Share |
|---|---|---|---|---|
| ACP | provider | 1,168 | 13 | 1% |
| ACP | client | 4,498 | 34 | 1% |
| Olas mech | agent | 1,573 | 5 | 0.3% |
| x402 | payer | 207,299 | 447 | 0.2% |
Each address in the month it first registered, against whether it has ever appeared on a rail since. The oldest cohort has had eight months and is at 0.2%; the best month reached 8.1%. Time is not the explanation.
Built by different people to different designs. Naming one of them “the” registry would be picking a winner on no evidence, so both are read.
| Registry | Identities | Held by | Declare a wallet | Feedback | Last touched |
|---|---|---|---|---|---|
| MPL Agent RegistryMetaplex's, attaching an agent identity to an MPL Core asset. | 1,587 | kept in the asset | — | — | 8 September 2026 |
| ERC-8004 on SolanaA port of the EVM standard, so the two sides are nominally the same registry. | 1,478 | 268 | 52 (4%) | 9,962 | 7 September 2026 |
3,103 accounts between them, against 329,320 addresses holding an EVM registration, and on the side that keeps a record, the same concentration the EVM registry shows at a five-hundredth of the size.
One keeps a record and one keeps a pointer. 8004 holds the holder, the operating wallet, the card and its own feedback counters in its account. MPL holds a reverse-lookup pointer and puts the payload in the Metaplex Core asset’s plugin, so its rows carry an asset and nothing else. Its columns are blank above rather than filled in from the other registry’s shape; following the pointer means decoding Core assets and their plugin registry, which is a separate build.
Counted 8 Sep 2026, 15:07. Recomputed daily; it is a full intersection over four tables rather than a sample.
| x402 | payee | 43,255 | 223 | 1% |
Read the other way round, this says the working addresses are mostly unregistered: the registry is not where the economy signs in. The 23 ACP jobs with a registered provider carry no recorded payment at all, against $1.03M paid across the book.
| April 2026 |
| 1,023 |
| Base66% |
| 67 |
7% |
| May 2026 | 1,731 | Base61% | 140 | 8% |
| June 2026 | 1,478 | Base46% | 49 | 3% |
| July 2026 | 6,170 | 30 | 0.5% |
| August 2026 | 284,040 | BSC93% | 263 | <0.1% |
| September 2026 | 14,603 | BSC74% | 27 | 0.2% |
Bars are drawn against the best month, not against 100%, or every one of them would be an empty track. Note the size of the August 2026 cohort: 86% of every address in this registry arrived in one month, 93% of them on BSC, and it has the lowest crossing rate of any of them. A crossing is only visible on the rails we read, so each figure is a floor.
The wallet column is the field that would join an identity to later Solana activity. Most declare nothing to join on. And the most-used card is one document behind 135 identities, which is the EVM registry’s shape too.
A refusal from an endpoint is recorded as a refusal, never as a registry with nothing in it.