Four of Every Five Dollars Flowing Into Listed x402 Wallets Settled Outside the Protocol

📅 Figures below are as measured 2026-08-03 (UTC), one ~22-hour window, a point-in-time snapshot published the day it was taken. This is a single day, measured on a Sunday into Monday; we will publish multi-day windows before treating the ratio as stable. For live numbers see the economy dashboard.

The mechanism is the receipt

An x402 payment is not a normal stablecoin transfer. In the protocol's exact scheme on Base, the buyer signs an authorization and a facilitator submits it on their behalf as an EIP-3009 authorized transfer: the calldata carries transferWithAuthorization or receiveWithAuthorization. A wallet that merely receives ordinary transfer() calls is receiving money, but it is not receiving x402 payments.

That difference is checkable by anyone, for any transaction, forever. So we checked our own catalog against it.

Controls first. The three highest-volume sellers in our catalog each sample 4 out of 4 recent settlements as authorized transfers: the fingerprint reads real x402 as real. A 25-wallet control slice measured 99.7% of dollars authorized, so the estimator does not smear clean wallets. And the reason we built the test at all: five wallets our discovery pipeline had flagged as promising candidates sampled 0 out of 16 authorized, fed at up to 19,000 transfers per day by a shared set of sender wallets. A farm can mimic micropayment sizes; it cannot mimic facilitator-mediated settlement without actually using a facilitator.

What the catalog looks like under the fingerprint

Population: all 1,092 Base wallets in our catalog that have ever been paid.

Bucket (22h window)	Wallets	Share
Dormant, no inbound stablecoin transfer in the window	825	76% of catalog
Active, settling on-protocol (≥50% of sampled txs authorized)	206	78% of active
Active, plain transfers only	51	19% of active
Active, mixed	7	3% of active

By wallet count, the protocol is healthy: most listed sellers that are receiving money at all are receiving it the way the spec says.

Then we weighed it in dollars

Group	Wallets	Inflow in window	Settled via EIP-3009
Plain and mixed wallets	58	$18,715	~$26 (0.1%)
On-protocol wallets	206	~$4,650	control slice: 99.7%

Total inflow to all active listed wallets in the window: about $23,400. Of that, roughly $4,700, about 20%, settled through the protocol. The average on-protocol wallet received about $23; the average plain-transfer wallet received about $320.

Read that again the way we had to: the x402 protocol's real economy is many wallets moving small money, exactly what agent micropayments should look like. The bulk of the dollars flowing into x402-listed wallets arrive as ordinary transfers that never touch the protocol the listing advertises.

The sender mesh

One more thing fell out of the sampling. Among the 58 plain-and-mixed wallets, 25 sender addresses each fund between 2 and 12 of them (the top sender funds 12, the next 10). Independent services do not share payers at that density. This is a coordinated network operating across marketplace-listed wallets, distinct from the quarantined farm cluster that prompted the test; the two share only one sender. We are feeding it through the same entity-clustering pipeline that produced our wash-trading findings.

What these numbers do not say

Plain-transfer inflow is not, per wallet, an accusation. An operator funding their own service wallet, an exchange withdrawal, a refund: all legitimate, all arrive as plain transfers. The finding is about the aggregate: dollars flowing into x402 listings overwhelmingly do not settle through x402. Which listings deserve stronger language is a question for entity clustering, per wallet, with receipts.
One day is one day. The window is ~22 hours. The wallet-count split has been stable across our samples; the dollar split we will only call stable after multi-day windows.
This is Base only, and Base is a minority of the dollars. Corrected the same day this posted, after a reader question: Base holds 81% of the seller wallets we track but only 37% of the measured dollars. Arbitrum (5 sellers) and Polygon (10 sellers) hold 57% between them, and this study did not look at either. We excluded them for having single-digit wallet counts, which is filtering on the wrong axis, and it is the same mistake this post warns about: a headline count hiding where the money actually sits. EIP-3009 works identically on those chains, so there is no technical reason for the gap. A nightly run now covers every EVM chain and we will publish the multi-chain figures rather than restating this one. Solana remains excluded for a real reason: it has no EIP-3009 and needs separate machinery, and folding it in would manufacture false 'plain transfer' verdicts.
Mechanism shares for the plain/mixed group are estimates from stratified samples of up to 30 transactions per wallet, extrapolated against each wallet's fully-counted total inflow. Sample sizes are recorded per wallet in the detail files.

Verify this

The study scripts (ops/fp_study.py, ops/fp_volume_study.py) are deterministic; rerunning them against the same window reproduces the same samples. Per-wallet detail files as of publication:

sha256(fp_study_detail_2026-08-03.json)        = 879d51bddd6ca5f5f3fd72ec236f3a8c73e6bbd5a9050fc0d70816be942a6013
sha256(fp_volume_study_2026-08-03.json)        = 1c29be61ede7b65aa3b4893c98720bb78f617a01b645488d0f11b3ad752f06a5

Every transaction behind these figures is public. Pick any wallet in our catalog, pull its recent inbound transfers on Basescan, and look at the calldata selector: 0xe3ee160e and 0xef55bec6 are the protocol; 0xa9059cbb is just money moving. Our standing figures remain sealed nightly in the integrity chain, witnessed off-site.

The Agent Almanac measures the agent economy at the settlement layer, on-chain, with the wash stripped out and the coverage disclosed. As of today, that includes measuring whether the money in "the x402 economy" actually settles through x402.