Measured 2026-09-10, on-chain, by two independent methods.
We published a census of the x402 economy on Base: every USDC transfer in a 24-hour window, joined against all 875 sellers in the public directory. It found 955 transfers and $312.12 across the whole catalogue, and it looked like a market where money moves in a few large bilateral payments.
Then a public dashboard showed a single service doing more transactions than our entire census. We checked, and it was right.
claw402.ai is an LLM gateway — 139 endpoints, an OpenAI-compatible API paid per call. It is not in the canonical x402 directory. We scanned all 14,200 records of that index: zero claw402 listings. Zero for Vishwa, another high-volume service. So we stopped reading indexes and asked the chain directly, scanning every USDC Transfer into claw402's payout address — 0xa9dd7cc9cbf0e05551332209289f04be36bc2315, taken from its own published openapi.json:
| claw402, measured on-chain | |
|---|---|
| blocks scanned | 2,000 (~67 minutes), 0 chunks failed |
| USDC transfers received | 329 |
| volume | $1.1780 |
| distinct payers | 10 |
| extrapolated to 24 hours | ~7,100 transfers, ~$25 |
Our entire 24-hour census found 955 transfers across all 875 listed sellers. One seller we could not see out-transacts everything we measured, by roughly 7x.
They are different economies. Split by population:
| transfers | volume | average ticket | shape | |
|---|---|---|---|---|
| Directory-listed sellers | 955 | $312.12 | $0.33 | few payers, larger payments |
| An unlisted high-volume seller | ~7,100 | ~$25 | $0.0036 | many small calls |
The "bilateral, few large payments" pattern we published is real — but it is a property of directory-listed sellers, not of the x402 economy. The many-small-calls economy is roughly an order of magnitude larger by transaction count, and it lives outside the directory entirely.
Not media, which is what we sell. Agent infrastructure: an LLM gateway, blockchain data, business-data enrichment, trading signals, machine banking. That is where the call volume is.
The population you can enumerate is not the population that matters. A directory is a list of who bothered to register, not a census of who is trading. Our mistake was not the arithmetic — the census counted correctly — it was presenting a measurement of a list as though it were a measurement of a market.
The original headline stays published, with this correction attached rather than quietly replaced. We would rather be the source that found its own error than the one that has to be told.
Method for both halves: seller addresses from a public directory pull for the census; for the claw402 measurement, eth_getLogs against the USDC contract on Base filtered on the recipient topic, in 100-block windows with the coverage stated and zero failed chunks. Re-runnable by anyone with a public RPC.
← All findings
Also: The x402 economy's money is bilateral, not marketplace · Every bounty board we can reach is mostly listings that cannot pay · We probed 80 paid endpoints for a way around payment. We found none.