How much x402 volume is real?

Why the headline numbers are not the answer#

x402 produces an unusually attractive number: every payment is on a public chain, so anyone can count them. The counting is easy, which is exactly why the result is unreliable.

Several things get counted as x402 volume that are not x402 commerce:

• Gasless transfers that are not x402 at all. The on-chain marker x402 leaves is shared with unrelated systems. Counting markers overstates the protocol.
• Self-payments. Payer and payee are the same party, or two wallets controlled by one. Nothing was bought.
• Circular flows. Money moving in loops between a small set of addresses, which inflates both transaction count and dollar volume.
• Scripted traffic. Real payments between real parties, where no agent was involved and no decision was made.

Each of these is cheap to produce and indistinguishable from commerce if you are only counting transfers.

What the four analyses found#

Note the figures do not agree with each other, and they are not supposed to. They measured different populations with different filters over different windows. What they agree on is the direction and the size of the correction: the honest number is much smaller than the headline, in every case, by every method.

How did they tell the difference?#

TRM's filters are the most clearly documented and they are worth knowing, because they are what any serious analysis has to do something equivalent to:

• Remove settlements where payer and payee are the same address.
• Remove flows concentrated in one or two payers, since a seller with one customer is not a market.
• Remove sellers with fewer than ten distinct buyers.

TRM also flagged the honest limitation of its own method, which most commentary left out. A single-purpose agent repeatedly buying one service looks exactly like a script. The filters that catch wash trading also catch some genuine automation, so the result is a lower bound on authenticity rather than a measurement of it.

That caveat is the difference between analysis and a headline. A filter that removes suspicious activity removes some real activity too, and a number produced that way should be reported as a range with its method attached.

What should you actually do with this?#

Three practical rules, whether you are evaluating the ecosystem or a single seller.

Never take a protocol-level total at face value. Reading a dashboard without being misled covers what to establish first, and part of the answer is that most assistants cannot pay at all, which constrains how much of this could be agentic. Ask what population it covers, what window, which chains, and what was excluded. A figure without those four is not a measurement.

Judge a seller by distinct buyers, not by volume. Volume is trivial to manufacture; a long tail of distinct counterparties is not. How to audit a service's traffic sets out the specific checks.

Separate the evidence grade from the count. A record that says "this many settlements, of which this many were relayed by a named facilitator" is far more useful than one big number. The second figure is the one you can stand behind.

Does this mean x402 is failing?#

No, and the conclusion people reach from these studies is usually too strong.

Two things are true at once. The headline figures are inflated, often by an order of magnitude or more. And there is real activity underneath them, growing, with identifiable sellers serving identifiable buyers.

The reason both are true is that the inflation is cheap and the genuine activity is small in dollars by construction. A protocol whose median payment is a few cents will have a small dollar total even when it is working perfectly. Measuring it by dollar volume, which is what a crypto-native analyst reaches for first, was always going to produce a disappointing number next to a manufactured one.

Transaction counts between distinct parties are the better signal, and they are the hardest to fake at scale.

Frequently asked questions#

Why would anyone wash-trade API payments?#

The same reasons as anywhere else: to appear in rankings, to qualify for incentives, to support a fundraise, or to make a protocol look busier than it is. The cost of manufacturing a payment to yourself is the gas, which on a cheap chain is close to nothing.

Can an index remove all of it?#

No, and one that claims to is overclaiming. Self-payment is detectable because the addresses match. Two wallets controlled by one party are not, unless they are careless. The honest position is to annotate what can be detected and say what cannot.

Is self-payment always illegitimate?#

Not always. Testing, internal routing and treasury movement all produce it. That is why annotating it is better than deleting it: a feed is a record of what happened, and the reader can decide whether to exclude it.

Which chains are worst affected?#

Reporting through 2026 found concentration problems on the chains with the cheapest transactions, which is what you would expect since cheap transactions make manufacturing volume cheap. Comparisons across chains should be per chain, never pooled.

Does Roundhouse publish a cleaned number?#

It publishes the evidence grade alongside the count rather than one adjusted figure, because the adjustment is a judgement and the reader should see what was applied. /stats states the basis and window for every figure on it.

Where to check this yourself#

Do not take this article's numbers either. /stats publishes the indexed record with its coverage and window stated, broken down by evidence grade so you can see the verified band separately from the total. /transactions lets you filter the raw settlements, and coverage and confidence documents exactly what each grade asserts and what it does not.