This content reflects minia2a as of its publication date and is kept for the record. Current model: x402 pay-per-call in USDC on Base only, 5 free trial calls per signed wallet, no credits and no top-up rail.
November 14, 2026 · measurements taken 11:44Z–11:49Z
Every week the supply-side numbers get bigger. Coinbase's agentic.market lists 4,242 services. CDP's Bazaar has settled 33,655 items. PayAI's facilitator serves 16,278 discovery resources. UltravioletaDAO serves 1,850. Our own catalog is at 1,893.
None of these numbers says whether anyone is paying. This week, for the first time, a public ledger does.
The one public demand-side number
Settled publishes a status endpoint that tracks x402 endpoints and records price-matching USDC payments against them. It is free and signed. We read it on 2026-11-14 at 11:49Z, and this is what it reports:
| Metric | Value |
|---|---|
| Endpoints tracked / probed | 175,450 / 175,331 |
| Live share of probed | 38.7% — 67,889 live, 98,965 dead |
| Sellers receiving a price-matching payment, Nov 7–13 (7 whole days) | 630 |
| Distinct payers in that window | 3,826 |
| Transfers / USD | 660,013 / $15,866.16 |
| Mean transfer | $0.024 |
| Scout purchases / spend / endpoints verified | 5,174 / $97.84 / 3,812 |
Method, stated plainly: these are Settled's own measurements from its own ledger, over seven whole days, filtered to payments whose amount matches the listing price. They are not a census of all x402 settlement — a seller whose price drifted, or whose payment settled outside the rails Settled watches, is not in the 630. Treat it as a floor on paying sellers and a fair estimate of the shape.
0.36%
630 sellers out of 175,450 tracked endpoints is 0.36%. One in 278 endpoints was paid, with a price-matching transfer, in a given week.
That is the number worth carrying. It does not mean 99.6% of endpoints are useless. It means the binding constraint on this market is not how many endpoints exist — it is how many endpoints sit in a path that a buying agent actually walks. Catalog size stopped being scarce a while ago. Placement in a demand path never stopped being scarce.
It also reframes the dead-endpoint rate: 61% of probed endpoints do not answer at all. Most of the 175,450 are not dormant inventory waiting for a buyer. They are gone. Tracking them inflates every supply-side count in the ecosystem, including the ones above.
A $0.024 mean transfer kills price as a selling point
660,013 transfers for $15,866.16 is a mean of 2.4 cents. At that size, a competitor being one cent cheaper is a rounding error against the cost of the agent discovering them, not discovering the alternative. Price competition on the catalog floor is not a strategy; it is noise.
What a buyer's agent spends instead is attention: finding an endpoint that answers, whose 402 is parseable, whose price matches what the catalog said, and whose response is the shape the schema promised. Every one of those steps is a place to lose the call. None of them is a place where shaving a cent helps.
What this changed in our own numbers
We ran the same lens over the indexes we can see ourselves:
| Where we appear | Rows | Why that many |
|---|---|---|
| x402scan | 145 of 145 demand-backed endpoints | Self-serve registration; 0 unregistered |
| agentic.market (Coinbase) | 5 endpoints | The index is a settlement snapshot — there is no submit form; only 5 of our endpoints have been settled through the CDP facilitator |
| CDP Bazaar | 56 items | Down from 98 on October 16. Listings age out when nobody pays for them for ~30 days |
The agentic.market row is the useful one. Coinbase's docs are explicit: "There is no registration form or separate API call." A row appears because a settlement happened. Our 5 rows are not a submission gap we can close by filling in a form; they are an accurate count of how many of our endpoints have had a completed CDP settlement. The Bazaar row going 98 → 56 over a month is the same measurement from the other side: listings are leases, renewed by payment.
That reframes the work. Submitting to more directories does not move those numbers. Getting an endpoint into a path where a buyer's agent will call and pay for it does — and a settlement is what proves it happened.
What we are not claiming. We do not know Settled's full sampling method, and 630 is their attribution, not an audited count. We do not know how much settlement happens outside the rails they watch. The live/dead split is a probe result at a moment in time, and probing methods differ. The one thing the data supports without qualification is the ratio's order of magnitude: paying sellers are a small fraction of listed endpoints, and the per-call amounts are cents.
The practical conclusion
- Stop counting catalog rows as progress. They measure supply, and supply is abundant. A row count is not a distribution metric.
- Treat listings as leases. Any index derived from settlement needs a payment inside its window or the row disappears. That is the index being honest.
- Optimize the call path, not the price. A parseable 402, an honest price, and a response matching the advertised schema are worth more than a cent off a two-cent call.
- Measure demand where it is observable. One public ledger with seven days of price-matching payments beats any number of supply-side totals, including our own.
The market is small and real. 630 sellers were paid last week. The interesting question is not who else can list an endpoint — it is which of those 630 got there by being in the path, and whether that path can be described well enough for another seller to join it.
Originally published at minia2a.uk.
