api.exa.ai/search

1.Catalog declaration

→ 0 more columns

2.Probe history

→ 0 more columns

3.L1 — real purchases

25 of 27 paid attempts settled with a receipt, and 10 of those also returned a 2xx response (delivered). Each settled row carries its on-chain transaction hash — the receipt is the evidence. A settled row reads as nonce-bound when the on-chain re-read also matched the one-time signature nonce we generated for that purchase, and amount + payee when it matched amount, payee, asset and chain with no nonce on record — the binding shipped on 2026-09-04, so earlier rows carry the weaker evidence and keep their label rather than being demoted (methodology). settled is the transfer we confirmed on-chain; delivered is the response arriving. A settled row whose paid request answered 5xx counts as settled and not as delivered — definitions.

17 of those paid attempts are held as inconclusive rather than counted against this seller (15 settled and answered 4xx, 0 answered 4xx with no settlement receipt, 2 answered 402 or 5xx while our own payer wallet was out of USDC, 2026-09-13 00:00 to 2026-09-15 23:49 UTC). A 4xx says the request was not one the server would accept, and we cannot rule out that the request was ours to get wrong: vet402 buys with no API key of the seller's, and sends a POST body only when the seller declares one. This is the same principle the methodology already applied to path_template URLs, applied to the body and the authentication header as well. The rows are not deleted and not hidden — they are below, with their status and HTTP code, and they are out of the denominator for delivered (corrections).

2 of the paid attempts failed on vet402's side (for example our wallet was out of USDC, or we did not send the input the listing declares). The rows stay in the table; the embeddable badge below leaves them out of its count and names them as “vet402 side”.

→ 0 more columns

[![vet402 receipt](https://vet402.com/api/badge/endpoint/521e929e-5f89-4603-a964-d1812caf118f.svg)](https://vet402.com/observatory/e/521e929e-5f89-4603-a964-d1812caf118f)

4.Catalog listing events

No listing changes observed.

unverified is not a failure; a fail is published only after two consecutive failing probes. Definitions: methodology.

5.Dispute this record

Think a measurement above is wrong? Say which probe or purchase and what you observed instead. One person reads it and replies. The record is never deleted on dispute: if it was wrong, the correction is published with the same weight; if it was right, it stands. Operators who control the receiving address can also sign a dispute via POST /api/v1/observatory/disputes (API reference), which re-measures through the normal publication gate.

About the purchase at 2026-09-25T00:02:17Z (UTC). The form starts with that time; add what you saw instead.