Week of 2026-09-10. Every number below was read from a chain, not from a dashboard. Raw data and the collectors that produced it are in this repository.
Most writing about the agent economy quotes somebody's self-reported counter. We count settlements. This is what that looks like on the two chains that carry practically all of it.
| Settled in 24h | $72.98 |
| Settlements | 3,685 |
| Distinct paying wallets | 76 |
| Median ticket | $0.002 |
| Services with more than one paying wallet | 4 of 23 |
Read straight from the Base USDC contract: every transfer whose recipient is a service's published payTo address. Self-reported counters are ignored entirely, and one of them is why we started doing this — a catalogue whose "sales" turned out to be 81% its own internal heartbeats.
Scope, stated up front because it matters: this is a census of the services in our catalogue, not of the x402 economy. Every service here was found by harvesting payment addresses from the public directory, so these totals are a floor on the market. DefiLlama's ecosystem-wide tracker put x402 settlement near $16,000 a day in July 2026, down from a ~$970,000 peak in December 2025 — roughly two hundred times the $73 below. The finding that survives that gap is not "the economy is tiny"; it is "most services in the catalogue have no customers", which is a statement about the catalogue and is exactly as damning as it sounds.
98% of it is one service — StableEnrich, at $71.32 of the $72.98. The rest of the directory is a display window with nobody behind the counter.
Corrected 2026-09-10: this section first said "5 of 23" and "two-thirds of the volume is one service". Both were wrong. The 5 counted our own service, whose only payments are the ones we make to test it — the same mistake that once leaked our own money into the totals, caught this time by the reviewer's question "which services?". The "two-thirds" figure came from an earlier snapshot and was never re-checked against the published data. The table above now matches data.json exactly.
The same question, asked of the other chain. Solana has no cheap log query, so this is a sample: every USDC-accepting seller address in the directory — 271 of them once the census was complete — signatures per address, then up to 25 transactions each.
| Sellers measured | 271 |
| With any on-chain history | 20 |
| Settled anything in 24h | 9 |
| USDC settled in the sample | $261.93 |
| Largest single seller | $237.59 — 91% of it |
The address that tops Solana's catalogue by listing count — 533 advertised resources — has never transacted on mainnet. Not once.
That $237.59 came from 13 settlements by 7 distinct wallets: an average ticket of $18.28 from a catalogue whose advertised median price is $0.02.
The individual payments are round: $25.02, $24.98, $24.97, $24.99, $27.03, and a few between $5 and $7. Either the address sells something its catalogue does not describe, or it receives payments that are not x402 calls at all. We cannot tell from outside, and we are not going to pretend we can.
What we can say is the rule it forces: an inflow into a payout address is not proof of a sale at the listed price. Our Base figures carry the same caveat, which is why the next version of the index will flag any seller whose realised ticket differs from its listed price by more than an order of magnitude instead of letting it top a leaderboard.
Issue 01 promised that the next version of the index would flag any seller whose realised ticket differs from its listed price by more than an order of magnitude. It is built, and the result is worse than expected.
Of the 271 Solana sellers measured, 9 settled anything at all in 24 hours, and 7 of those 9 are flagged. (The other two transacted within a 10x band of their listing, which is what an honest catalogue looks like.)
| Realised median ticket | Listed median price | Ratio |
|---|---|---|
| $14.7485 | $0.001 | 14,748x |
| $7.8285 | $0.002 | 3,914x |
| $24.9839 | $0.02 | 1,249x |
| $0.0001 | $0.02 | 0.01x |
| $0.0001 | $0.01 | 0.01x |
| $0.0001 | $0.005 | 0.02x |
| $0.0001 | $0.004591 | 0.02x |
Seven of the nine sellers that took money at all did so at a price their catalogue does not describe. Three took between a thousand and fifteen thousand times the listed price; four took around a hundredth of it.
This is the finding behind the finding. The directory is a price list, and on Solana the price list does not describe the payments. Anyone sizing this market from catalogue prices — including an earlier draft of this report — is reading fiction. It is also a caution about our own numbers: a payTo address is a wallet, and money arriving in it is not proof of anything except that money arrived.
Pulled in full — and "in full" took three attempts, which is itself the finding:
| Resources | Distinct sellers | |
|---|---|---|
| Base | 23,899 | 879 |
| Solana | 4,367 | 289 |
The discovery endpoint paginates and gives no total. Draft one read its default first page of 100 and concluded Solana had overtaken Base (485 accepts against 442). Draft two paged to 7,000 resources and published Base 5,457 / Solana 1,534. This version paged until the endpoint returned nothing and found 28,377 resources. Each correction moved the same way: the directory was bigger, and Base's share of it larger, than the previous look suggested.
The first two drafts are left described here rather than deleted, because the failure was not a typo. It was stopping when the data looked like an answer.
Everything above says the same thing — catalogues that nobody uses. So here is the measurement that says the opposite, and it is the reason this section exists.
Every tokenized equity on Solana ("xStocks") — all 74 of them, not a sample of the big ones:
| Combined market cap | $1,002,719,270 |
| Combined holders | 416,982 |
| Signatures touching those mints in 24h | 256,809 (a floor — 20 of 74 hit the read cap) |
| Estimated value moved per day | ~$66.2M |
| Median token's movement in the sample | $0 |
| Tokens moving more than $1,000/day | 19 of 74 |
| Largest token's share of the total | 71% (STRCx) |
| Top three tokens | 86% |
So the honest version of the counter-example is not "tokenized stocks work". It is: tokenized stocks work for about a quarter of the tokens, and barely at all for the rest. 416,982 wallets hold these things; 55 of the 74 tokens moved under $1,000 in a day, and the median moved nothing at all in the transactions sampled.
And the aggregate should be read loosely. The estimate comes from a uniform sample of 40 signatures per mint; STRCx alone implies $47M/day from 27 transfers caught in its sample, so the total's confidence interval is wide. The shape — a long dormant tail and three tokens carrying nearly everything — is the robust part, and it is the same shape as the x402 directory: presence is cheap, flow is not.
How this number was very nearly wrong. The first version of this measurement read the newest 30 transactions of each mint and reported their volume as the day's activity. Across the twelve largest that gave $122,000 — and it would have gone out as "NVDAx moved $800 today" against a token with 80,418 holders. Paging the signatures afterwards showed 9,000+ transactions per mint in 24 hours: the sample had covered 0.3% of the window, taken entirely from its head, and understated the flow by roughly 45x. Newest-first is not a sample — on a chain, the newest transactions are whatever just happened. The figures above come from a uniform sample across the whole window, with the sampling fraction stated, and they are labelled estimates.
What this means for the rest of the report. The x402 directory's emptiness is a fact about that directory, not about the agent economy, and it would have been easy — and wrong — to generalise from it. Two markets, measured the same way, one empty and one turning over millions a day.
issues in repositories over 20 stars: 30% were already paid and left open, 59% pay in project tokens rather than money, and zero were viable. The largest "available" listing has 67 prior attempts and a 0% merge rate.
published "earn" programme. Zero advertised one. The directory is a directory of things that charge.
Settlements are counted from chain data: ERC-20 Transfer events on Base, SPL token balances on Solana. A payTo address is a wallet, and a wallet can receive money for reasons that have nothing to do with the catalogue it is attached to — see the $237 above. Volume is inbound only. The Solana figures are a sample of 271 sellers — every USDC-accepting address in the directory, with up to 25 transactions read per address — and are labelled as such everywhere they appear.
A second correction, 2026-09-10: this issue first reported 95 Solana sellers, 16 with any on-chain history and 7 with a settlement. Those came from a seller list built on a truncated directory pull (see the table above — 1,534 resources instead of 4,367). The figures here now come from the complete list. The price-mismatch finding is unchanged in shape and got slightly stronger: 7 of 9 rather than 7 of 7, with two sellers landing inside a normal band.
Everything here is re-runnable from the repository, including the parts that make us look bad. That is the point.