# The Assay: Who Actually Pays x402 Sellers

October 2026. ProvenPaid market note no. 2, by Justin Smith, Ph.D. Measured 2026-10-18 from public Base data. Previous note: https://api.provenpaid.com/notes/state-of-x402-demand

In a gold rush, the assay office is where claims meet the scale. A miner can say the vein is rich; the assayer weighs what came out of the ground. This note weighs x402 sellers the same way: not what their listings say, but who paid them on-chain.

Four findings:

- **Most listings never reach a paying wallet.** 202 projects in the x402.org directory narrowed to 41 live x402 endpoints, which pay into just 6 distinct receiving addresses on Base.
- **A valid paywall says nothing about demand.** Two of those six sellers answered with a valid x402 challenge every day we checked, and received no x402 payments at all in 30 days.
- **Real demand exists, and it is concentrated.** The busiest seller took about 3,200 payments a day from 198 wallets, but its top five payers made three quarters of them.
- **Independent checks catch what sellers' own books miss.** A seller asked ProvenPaid to check its revenue against the chain. The chain showed more settlements than the seller's own counters.

## From 202 listings to 3 sellers with customers

![202 listings narrow to 6 receiving addresses and 3 sellers with more than two payers](/brand/notes/seller-funnel.svg)

We started from the public x402.org ecosystem directory: 202 projects. For each, ProvenPaid sent unpaid requests to its website and to its published manifest, if any, honouring robots.txt. That found 41 live endpoints answering with an x402 payment challenge; 34 were valid on October 18. ProvenPaid has probed them daily since October 12.

Those 41 endpoints pay into only 6 distinct receiving addresses in USDC on Base. One seller alone runs 10 endpoints on one address. So "dozens of x402 services" often means a handful of businesses.

This is the slice we can find without proprietary data. Explorers that see facilitator traffic count far more sellers. The point is not the absolute number; it is how quickly a long list of listings becomes a short list of businesses.

## What the six weighed

Every figure below is from public Base USDC logs over the 30 days to October 18, using the same method as a ProvenPaid check: one settlement per EIP-3009 authorization, payers who were paid back by the seller set aside. Sellers are anonymised; this note reports evidence, not ratings.

![Share of each seller's x402 payments made by its top payer, its next four payers and everyone else](/brand/notes/payer-shares.svg)

| Seller | Payers | Returning payers | x402 payments | USDC received | Median payment (USDC) |
|---|---|---|---|---|---|
| A* | 198 | 85 (43%) | 23,432 | 375.987 | 0.0059 |
| B | 60 | 15 (25%) | 180 | 0.528 | 0.0020 |
| C | 12 | 11 (92%) | 107 | 2.290 | 0.0100 |
| D | 2 | 0 (0%) | 2 | 0.015 | 0.0075 |

Sellers E and F received no x402 payments in the 30 days. *Seller A: about 7 of the 30 days scanned (October 11 to 18), because its volume reached the node provider's limit.

Seller A is so busy that the scan reached its provider limit after about seven days of history (October 11 to 18). Even that slice is 376 USDC, roughly 3,200 payments a day at well under a cent each.

Three patterns stand out:

- **Volume and breadth are different things.** Seller A has both: 198 paying wallets. But a third of its payments come from one wallet, and three quarters from five. That is the shape of a few automated buyers doing most of the work, the same shape the first note found ecosystem-wide.
- **Loyal but small is a real business too.** Seller C has only 12 payers, but 11 of them came back more than once. Repeat business is the signal a listing cannot fake cheaply.
- **Live is not the same as used.** Sellers E and F passed every protocol check, every day. An agent that only checked "is this endpoint up and well formed?" would treat them exactly like Seller A.

## A seller asked to be checked

The operator of the Automaton-Sovereign service on Base asked ProvenPaid to check its revenue against the chain, and published their own counters for comparison. The thread is public: https://github.com/justinsmithphd/provenpaid-mcp/issues/3

Two things came out of it:

- **The chain disagreed with the seller's books.** In the period their counters cover, ProvenPaid counted 58 on-chain settlements from outside payers against 47 recorded settled calls, plus 60 more from wallets the seller also paid back. Batching would push the on-chain count below the call count, not above it, so the gap points at the seller's counter, which the operator had already flagged as unverified. By October 18 the address showed 144 settlements from 6 payers, with one wallet making 94% of them.
- **They made our method better.** The operator contributed a test fixture of ten tricky settlement patterns, including batches, retries, relayers and self-payments. It caught one counting rule in ProvenPaid, which we fixed: five authorizations in one transaction are five payments, not one. All ten cases now run on every change.

That is the job in miniature. A seller's own dashboard is a claim. The chain is the scale.

## What this means for agents and the people who build them

- **Check payers, not just paywalls.** A valid x402 challenge is necessary, not sufficient. Two of six live sellers here had no paying customers.
- **Look at concentration.** Hundreds of payers can still be a few heavy wallets. Top-five share tells you more than a raw count.
- **Repeat payers are the cheapest honest signal.** A wallet that pays twice probably got something worth paying for.
- **Expect wash and self-payment to be invisible to logs alone.** A seller paying itself from a second wallet looks like a customer. Concentration is the clue, and a seller's willingness to be checked is evidence too.

## The assayers are multiplying

Trust tooling for x402 keeps arriving ahead of the buyers who need it. Since the first note, new monitors probe ProvenPaid every day; on October 18 automated clients asked our price about 550 times and paid none. Other services now offer pre-payment checks too, including x402station, which publishes an open signal format and a free trial, and Endpoint Risk Corpus, listed on EntRoute. We describe them only from their own public pages.

The first note argued the town has more shops than miners. The scale says the same thing more precisely: a few sellers with real customers, a long tail with none, and a small number of wallets doing most of the paying.

## Work with us
- **Platforms, directories and routers:** want these figures for every seller you list, refreshed daily? Write to contact@advancedanalyticsgroup.org.
- **x402 sellers:** want your own payer depth independently verified, the way the Automaton check worked? Your first report is free. Write to contact@advancedanalyticsgroup.org with your endpoint URL.
- **Agents and builders:** run a check before paying an unfamiliar endpoint. Three ways to try it:
    - From Claude or another MCP client: `uvx provenpaid-mcp` (free checks, no funds needed).
    - From code: `GET https://api.provenpaid.com/v1/check?url=<endpoint URL>`, $0.01 over x402.
    - See what a result looks like first: https://api.provenpaid.com/sample

## Method and limits

- **Sources:** x402.org ecosystem directory (2026-10-09 snapshot); ProvenPaid unpaid probes and its daily seller watch; USDC Transfer and AuthorizationUsed logs and block timestamps on Base via a commercial node provider. No Coinbase Bazaar or CDP discovery data is used.
- **Counting:** one settlement per EIP-3009 authorization; other inbound USDC reported separately; payers the seller also paid in the window set aside as round-trip payers. Full method: https://api.provenpaid.com/methodology
- **Limits:** a receiving address can serve several endpoints and other businesses; Seller A's figures cover about seven of the 30 days; the directory is one source among many, so this is a sample, not a census; a seller paying itself from a second wallet cannot be detected from logs.
- **Disclosure:** ProvenPaid sells these checks. We have no payment relationship with any seller named or anonymised here.