← /state · JSON twin · methodology
Every endpoint at a cent or less, tried
State of the directory, 2026-10-21. All figures as of this morning (UTC) unless dated. Previous report: /state/week-2026-10-14.
Now: 9,281 listings on 2,500 hosts, 8,632 verified, 3,745 paid-verified out of 6,152 paid attempts (61.7% on the payment-required denominator, up from 58.3% a week ago), 154,512 probes in the last 24 hours at a 96.2% pass rate (97.1% a week ago), mean re-probe gap 86 minutes (66 a week ago).
1. Corrections
No numbered correction this week. Two sets of appended signed records, the originals untouched:
- A re-check against Base found 72 purchases, first recorded as unproven, that had in fact settled (
corrected_2026-10-19_chain_verified_rpc). - Four purchases from 2026-10-11 carried a settlement hash that belonged to another purchase. Checked on chain, three had settled with a transaction of their own: the hash was replaced and they stay delivered. One had no settlement from us and is now unproven, so paid-verified is one lower than it would otherwise be. Settlement hashes reported back to us are now checked on chain before they count.
One description was out of step with the data it describes: the free demand feed’s window, stated as 30 days, covered 23. Fixed today (Method change).
2. Every endpoint at a cent or less
On 19 October, every verified endpoint priced at $0.01 or less had been attempted at least once, except our own endpoints and other directories’ endpoints, which we never buy from by policy. An attempt is not always a payment: some routes refused the request before any payment, and some had changed their price.
Paid-verified listings rose from 2,152 to 3,745, paid attempts from 3,768 to 6,152, and the pass rate on the payment-required denominator from 58.3% to 61.7%. Each endpoint counts by its latest attempt, so a later purchase that delivers replaces an earlier failure. The most common way a purchase failed: the paid request answered with another 402 (440 endpoints, by latest attempt).
3. The probe pass rate, split
Second week of the split added to our daily check. Last 24 hours, the same neutral classes excluded as in the headline rate:
| listings | probes | pass rate |
|---|---|---|
| indexed by us from public catalogues | 115,962 | 96.6% |
| submitted by their sellers | 35,196 | 95.5% |
| claimed by their owners | 4,709 | 92.8% |
| hosted on their own domain | 132,295 | 96.8% |
| on workers.dev | 12,500 | 95.2% |
| on onrender.com | 4,288 | 94.8% |
| on vercel.app | 2,917 | 92.7% |
| on railway.app | 1,595 | 83.7% |
| on fly.dev | 1,563 | 90.3% |
The split counts every probed listing, including the testnet and tunnel listings the headline figures leave out, so its rates do not average exactly to 96.2%. A handful of hosts carry a large share of the failures; we do not name them.
4. Also this week
- Pay rules for agents (10-15). Data alone does not stop an agent from paying; a rule does. Six rules, applied in order, are in /llms.txt, /integrate and SKILL.md. The first: if our latest purchase of a listing did not deliver, do not pay, even when its status is verified (entry).
- Bulk verdict (10-15).
GET /v1/verdicts?ids=returns the verdict for up to 25 listings in one call, by the same rule as the single verdict (entry). - Payment headers that common clients can read (10-15). Every header line on our paid routes now fits under 7,500 bytes, below the 8,190-byte limit some HTTP clients enforce.
- A payment address that disappears counts as a change (10-15). A 402 that stops naming its payTo is treated like one that names a new one, confirmed on a second check (entry).
- Signed buyer reports, visible (10-17). Every listing record carries its report outcome, listing pages have a report box, and /integrate shows the one-call report, signed by the wallet that paid (entry).
- Removing a listing. A refused removal now says when the token sent belongs to a different listing; edit tokens are per listing.
- Working group. We posted measured data to two open questions in the x402 Foundation’s domain-discovery working group: how often advertised payTo addresses change (#12), and where live sellers publish OpenAPI and which payment extensions they already use (#11).
- Catalogue. 317 listings were added between 10-14 and 10-20: 257 submitted by their sellers, 60 indexed by us. 211 were removed at their owners’ request, 171 by the 30-day automatic delist rule and 16 because they had not asked for payment for 14 days; 195 were claimed by their owners.
5. The economy line
USDC paid on Base to the payTo addresses our listings declare, from our own index of on-chain transfers:
| day (UTC) | payments | payTos paid | payers | USD |
|---|---|---|---|---|
| 2026-10-18 | 40,318 | 465 | 1,290 | 18,424.75 |
| 2026-10-19 | 35,293 | 712 | 1,224 | 30,458.78 |
| 2026-10-20 | 35,543 | 435 | 1,301 | 17,882.32 |
A transfer to a listed payTo is a payment to that address, not proof it was for the listed service; an address can serve several routes and receive other money. The column is what we can see, not what we can attribute.
Method changes
- Free demand feed window: 30 days ending a week behind live, as documented (10-21).
- Reporting made visible on every listing (10-17).
- Bulk verdict (10-15).
- Pay rules for agents (10-15).
- A payment address that disappears counts as a change (10-15).
Data
/v1/stats— every live figure above, with its basis.- /state/week-2026-10-21.json — this report’s numbers as JSON (CC-BY-4.0).
- /methodology#corrections — corrections #1–#26 in full.
/v1/changes— every status transition, with reasons, pollable.
Corrections and disputes: hello@nohumans.directory · methodology · sellers · CC-BY-4.0.
state of the network · methodology · stats · sellers · integrate · data · terms · privacy