992 x402 observations: why header and body checks must stay separate
992 x402 observations show why header and body checks must stay separate
CSOAI / Council of AI | Analysis published 21 September 2026
For x402 builders: do not turn a historical response-body predicate into a verdict about the whole protocol. In our published 21 September snapshot, 992 responses carried the recorded v2-plus-Bazaar conditions in the header while failing the dataset's narrower historical body predicate. That is a difference between instruments, not evidence that those 992 services are broken.
What the stored rows actually say
We recomputed the following from the exact dataset revision below. The observation ended at 2026-09-21T03:05:36Z; this analysis does not refresh that observation.
| Recorded property | Host rows |
|---|---|
| Distinct hosts, one stored probe per host | 2,636 |
| Answered HTTP 402 | 1,635 |
| Matched the historical body-based predicate | 479 |
| HTTP 402 with recorded header x402Version=2 and a Bazaar object | 1,467 |
| Header conditions above, but not the historical body predicate | 992 |
| No HTTP response recorded | 22 |
These categories overlap. They are not a funnel and must not be added together. In particular, neither 479 nor 1,467 is a count of verified paid deliveries, compliant implementations, unique businesses or customers.
The existing conformant field preserves the historical instrument: HTTP 402, a PAYMENT-REQUIRED header, x402Version=2 in the body, and a Bazaar object in the body. The separate header fields record another observation. This note does not silently rewrite either field or declare a new benchmark winner.
Why it matters when you build a client or a directory
The official v2 HTTP transport places payment requirements in PAYMENT-REQUIRED; the response body is implementation-specific. The protocol documentation also separates payment requirements, a client's payment payload and the server's settlement response. A settlement-response header can describe failure. Its presence alone is not settlement success.
A useful inspection record therefore separates the advertised endpoint, observed challenge, offer interpretation, authorised payment, settlement outcome and delivered artifact. Check each only when the required evidence exists. A free challenge observation need not and should not manufacture a payment to improve a ranking. Unsupported or unavailable evidence remains a visible limitation.
Exact source and scope
Dataset revision: bb38ad216f8e065c966587d0a8cc2d40647b83b1.
- summary-2026-09-21.json - SHA-256
d59f06e8c3221a25bd84d0ac51482c355911b041cfee02494ab29fc0468f1486. - snapshots/conformance-2026-09-21.jsonl - SHA-256
4b7f4aae533648f15adffc4aa26f930846051b19ea4313723a3023a0d62151dd.
This is analysis of existing public records, not a new crawl. The producer reports completed index reads; offset-based enumeration is not a transactional snapshot of every service. One URL per host does not represent every route. No payment, signed request, settlement or delivered paid artifact was tested. Missing HTTP responses do not establish permanent outages.
A useful next step
Maintainers can use the pinned rows to examine the observation attributed to their service, then share a dated, redacted correction in this thread. Please do not post credentials, customer data or payment signatures. We retain the historical record and distinguish a correction from a new measurement.
For scoped evidence-maintenance enquiries, use the Council of AI contact page. Public source inspection and verification are separate from any commissioned assembly or monitoring work. No favourable verdict is sold.
Published by CSOAI, the dataset producer. This is not independent validation, a security grade, a reserve audit, investment advice or protocol endorsement.