x402's First Honest Adoption Report: $15M Real, ~90% of Activity Filtered as Wash-Trading

This week the x402 agent-payment protocol got something it has never had before: an independent, honest accounting. Visa and Artemis published a joint report — Agentic Payments from the Ground Up — that is the first third-party attempt to measure how much of the agent economy is real. The headline numbers are modest, but the methodology is the story.

The numbers, with their own correction built in

The report counts x402 — the HTTP 402 Payment Required protocol for machine-to-machine micropayments — in its first year of production (the protocol launched May 2025). Two sets of figures matter, and the gap between them is the finding:

Read the two columns carefully: the report filtered roughly 90% of the raw dollar volume out before publishing the adjusted figure — classifying it as wash-trading, test transactions, or non-economic activity. Activity concentrates on Base, Solana, and Polygon, and settlement is almost entirely USDC.

That filter is not a footnote. It is the most important number in the report.

Why the 90% matters more than the $15M

A young financial protocol's raw on-chain footprint is almost always inflated — bots, testnets, self-trades, incentive farming. What makes this report unusual is that a major payment network and an analytics firm said so, in public, with a number: only about a tenth of the raw activity survived their honesty filter.

This matches what the settlement side has been saying. Circle reported on its Q2 earnings call that 99.3% of x402 agent-payment volume settled in USDC — a signal that the marginal transaction is real and tiny, not a large speculative transfer. It also matches what independent trackers show: raw 30-day volume numbers that are an order of magnitude below the adjusted annual figure, because the honest, economic slice of this market is still small.

None of this is bad news. It is the difference between a hype number and a usable one. $15 million of real machine-to-machine payment volume in year one — with 422,000 distinct buyers and a few thousand sellers actually charging for calls — is a real market forming. The 90% that got filtered out is the noise that will always attach to any new on-chain economy, and the report's value is that someone finally separated the two.

Signal vs. noise is the actual frontier

The protocol rails are settling. Settlement is commoditizing to USDC. The rails are no longer the interesting part — which is exactly why this report, and not a launch announcement, is the week's meaningful event. The interesting problem is now verification: given that 90% of what's out there is wash-trading and tests, how does a buyer tell a real, enforcing, delivering endpoint from noise?

We have been running this experiment from the other end. Our black-box payment-verification probe makes adversarial requests against live x402 endpoints — a forged transaction hash, a forged signature, a replayed payment — and checks whether the endpoint delivers results for free or rejects the fake proof. Across the 150 highest-traffic hosts in the public discovery directory, the answer so far is reassuring on one axis: zero free-riding. Nobody shipped a result to a forged payment. The verification layer is holding up better than the wash-trading stat might suggest.

But note the honest boundary. "No free-riding" and "real volume" are different questions. Our probe tests whether a seller verifies payment; the Visa–Artemis filter tests whether the activity itself is economic. Both point the same direction: in an economy where the raw footprint is 90% noise, the scarce resource is not another payment rail — it is a way to tell signal from noise before you pay.