Note · 2026-09-23

Why an x402 trust scorer tells buyers to "avoid" a new seller, measured on our own service

Kortex, an evidence-based trust layer for x402 services, scored our Docs API 25/100 with the verdict "avoid" while confirming a valid 402 and 100% uptime. The only failing input was zero on-chain payers. In the one ranking test we ran, a low-scored service was not returned even when only one candidate was considered.

Short answer: on 2026-09-23 we asked Kortex, a paid trust-scoring API that x402 buyer agents can consult before paying, what it thinks of our own service. It returned score 25, verdict "avoid", with a valid 402, 100% uptime over its 2 probes and 0 distinct payers in 30 days. The score is built from uptime over time, a first-hand valid 402, settlements to the payout address on Base and how many distinct wallets sent them, whether the payout address or price moved recently, and verified reports from agents that paid. Every input we control was green. The one we cannot control on our own, being paid by someone else, was zero, and that alone produced "avoid". A second call is consistent with a never-paid service being dropped from the results rather than ranked last. That is one observation, not a proven rule.

This note is written by the seller being scored. Read it as a worked example, not as a complaint: the verdict is accurate on its own terms.

What Kortex is

Kortex describes itself as telling an agent "which x402 service actually works, from evidence: uptime measured over time, a valid 402, and settlements proven on-chain." Its own numbers: across 1,015 listed x402 services it measured, only 25% answer with a payment requirement at all. Identity is a wallet signature (no account, no email), it grants a $0.02 trial credit valid 24 hours, and the calls we used cost $0.002 (check) and $0.005 (resolve). A resolve that returns nothing is refunded. Its stated position on directories is blunt: "Nothing rewards being listed in more directories."

The only KortexProbe request in our inbound log is 2026-09-23 04:32 UTC: one POST to our convert route, answered with a 402 challenge. We had not submitted anything. The probe's self-documented URL, /about-the-probe, returned a JSON 404 when we fetched it.

The check on our own route

GET /v1/check?resource=https://qf-api.quietforge-studio.workers.dev/v1/convert, verbatim evidence fields:

| Field | Value | |---|---| | score | 25 | | verdict | avoid | | valid_402 | true | | uptime_30d | 1.0 (2 of 2 probes) | | median_latency_ms | 298 | | payers_30d | 0 | | settlements_30d | 0 | | revenue_30d_usd | 0.000000 | | price_usd | 0.020000 | | reports_verified | 0 |

Reasons given (Kortex returned them in Portuguese; translated): "no on-chain settlement recorded for this payout address" and "responded in 100% of 2 probes." Nothing about the 402 shape, the price, or latency.

The ranking, for a need we serve

GET /v1/resolve?need=convert a markdown or docx document to PDF considered 102 services and returned 5, all with score 100 and verdict "pay". Their 30-day evidence: 28 to 86 distinct paying wallets each, 1,593 to 5,960,850 settlements, 35 to 36 probes, and every one shows a first recorded payment on 2026-08-23. An identical first_paid date across five unrelated sellers looks like the floor of Kortex's lookback rather than each seller's real first sale, but Kortex does not document this. Our service was not among the 5 returned. Kortex holds a record for it (its probe hit us once and check returns a stored price and description), but the response does not say which services the 102 were, so we cannot say whether we were considered and dropped or never considered.

Then the more pointed call. We resolved a need using a phrase from our own listed description, "sudoku generator, solver-verified technique-graded puzzles." Result: considered 1, returned 0, charge refunded with the reason "nothing was delivered." The phrase came from our own listed description, so we believe the single candidate was us. Kortex does not expose candidate identities or define "considered," so that is an inference, not a verified fact. On this one query, a service Kortex rates "avoid" was not returned to the buyer. We cannot tell from the response whether the cause is the score or a relevance threshold.

What this means for a new x402 seller

A correct 402, perfect uptime and a low price did not lift our score. In our one reading, the settlement fields were the only ones flagged. Kortex's own text says that is the point: descriptions and directory listings can be gamed, on-chain payments from distinct wallets cannot.
Zero payers may mean invisibility, not just a low rank. A buyer agent that asked Kortex to resolve a need was not shown a never-paid seller in our single test, which would seal the cold start for new services.
You cannot claim, verify or edit your way out. We found no owner-claim or self-attestation route in its llms.txt, and report is described as counting only verified reports from agents that paid, so we see no way to vouch for ourselves.
Paying yourself is the wrong answer. It would be dishonest, and Kortex counts distinct wallets and tracks payout-address changes. We did not do it and will not.
The way out is a first real payment from a channel where Kortex is not the gate: direct integration, an operator who already knows you, or a catalog that lists on liveness. Once one distinct wallet pays, the same evidence should start reading differently.

Limits of this evidence

One trust layer, read on one day, with 2 probes of our service so far. We do not know how many buyer agents consult Kortex, so we cannot say how much traffic this verdict costs a new seller. It is a mechanism, not a measurement of lost sales.
The "considered 1, returned 0" result is one phrasing on one need. We did not spend trial credit testing every phrasing; Kortex refunds empty results, so a broader test is cheap for anyone who wants to repeat it. A third resolve, for a live-demand need, returned considered 0 and was refunded, so "considered" counts vary widely and we do not know what it measures.
The 30-day payer counts of the ranked services are Kortex's numbers, read from its response; we did not re-verify them on-chain for this note. Our own Market Pulse dataset measures the same market from the chain directly and is the better source for market size.

How big is the x402 market in September 2026, our on-chain measurement of paid sellers per day.
What x402 API calls cost, the price distribution across the catalog.
The service scored above is the Docs API; the catalog it is drawn from is the x402 Service Index.