Where an agent’s money comes from

Agents by the share of their money that comes from one address. The last band holds the ones taking all of it from one.

A high share is not a finding against an agent. One large customer is how most businesses start, and an agent built for one buyer is doing what it was built for. What it IS, is the thing a counterparty count hides: 973 of these have exactly one payer, and the rail they settle on reports them beside an agent with four thousand.

The whole chain, and where it breaks

A payment proves money moved. Four things together would prove an agent is a working business: somebody paid it, it bought or did something, the result was recorded as delivered, and the customer came back. Each link is counted over the same 3,159 addresses.

1. 1A customer paid it3,07297%

   Nearly always showable, and on its own it proves only that money moved.

2. 2It ordered or did work43614%

   The link that breaks first, and it is OUR blind spot: an agent that buys nothing and one that buys from a wallet we cannot connect to it look identical here.

3. 3The result was delivered68422%

   Only ACP and the mech marketplace record completion. x402 is most of the volume and records payment, never delivery.

4. 4The customer came back2,43177%

   Repetition is evidence of a working service and not proof of one: a loop that pays itself repeats too.

What the record reaches

Each share is over the same 3,159 agents, so they can be read against one another. The gaps are ours and are the reason three of the six questions about a typical agent come back unmeasured.

The registry and the economy barely meet

3,073 addresses have been paid and never declared anything. 7 declare a service and 4 of those have an endpoint that answers; 16 carry a name of any kind. On this evidence the thing that declares an identity and the thing that earns money are different populations that barely overlap, which is why a reputation score built on registrations describes almost nobody who is paid. It is also why we cannot connect a second address for most agents: the link needs either a declaration these addresses do not make, or money moving between them. Funding from a common source is the third route and we refuse it, because an exchange funds thousands of strangers.

Asking this about one agent

The page above is the population. Six questions sit behind it: are the addresses paying it independent, is revenue growing, how much is repeat, how many payments bought anything, what does it cost to run, and can its identities be linked. They are answered for one agent at a time by agent_business, and what several agents rest on at once by agent_dependencies. Both are priced per call with no account. The counts they are built from are free on counterparty_check, and what is sold is the assembly and the four refusals: an address is not a customer, a payment is not a delivery, a shared funder is not an identity, and a cost we cannot see is not a cost that is not there. You can run our own cases against us before believing any of it. For the other concentration, how few addresses do the paying across the whole rail rather than inside one agent’s revenue, see who actually pays.