Half the x402 catalog never answers. Most of the other half has never been paid.

x402 is the HTTP-native way for software to pay for a web resource. A human with a browser or today mostly agents (we call them buyer) ask for a webpage or information behind an API, i.e. accesses a URL (an endpoint), gets back a 402 Payment Required with the terms. Then pays in a stablecoin, and retries with proof. The payment happens without an account, subscription, or checkout. The operator behind one or more endpoints is a seller; Some sellers run thousands of endpoints. On the blockchain we can only observe the wallets of the seller and payer. A buyer can have multiple payer wallets. Most larger sellers have multiple wallets as well even on the same chain. This sounds complex and it is.

Let us put ourselves in the shoes of an agent buyer. Before making a request to an endpoint and spending real money, she will ask herself four questions:

First Question: Which services = endpoints are available?

To the rescue: There are catalogs available like Bazaar, PayAI and others that advertise more than 150,000 x402 endpoints and growing. But are those reliable?

Over the last 30 days, we at Agents Trust probed those endpoints. We sent daily payment requests. Only 35% answered every time. More than half never answered once. Catalogs accumulate stale entries. A listing says a URL is offered; it does not say the URL answers today. Figure 1 captures our analysis of the x402 ecosystem.

Second Question: Which services are really used in practice?

Unfortunately, usage of individual services is not public. However, seller and buyer wallets are. We have indexed the x402 settlements on Base, Solana, Polygon, Algorand, Tempo and a few other chains with limited x402 activity today like Arbitrum, Optimism and World. Among the 5,660 sellers with catalogued endpoints, 55% have a live endpoint, 45% do not. The median seller has only half of its advertised endpoints live, answering x402 request.

Of the 3,139 sellers with at least one live endpoint, 23% paid in the last 30 days, 4% paid before, 73% never paid.

Third Question: Which endpoints are recommended to fulfill the needs?

But how can the buyer find out which endpoint does what? HTTP Payment Request headers according to the x402 standard, carry a description field and an output schema field. Those fields are then also advertised in the endpoint catalogs and made searchable. But, here again, the buyer is faced with a major gap. We found out that only 78% of the 150,000 plus advertised endpoints in the catalogs carry a “description” and only 23% have both description and output schema published. Also, most descriptions are not specific and meaningful. This is why Agents Trust offers a recommend API that helps agents and humans to find the right endpoints based on their specific need.

Recommendation is very important because there are competing endpoints with very different conditions. For illustration, take the need="weather forecast". We have identified 162 endpoints that offer weather forecasts. The prices vary 54-fold for basically the same information; most wrap a similar third party service at differing cost. The request schema and scope differ strongly as well. Some offer only results by ZIP code in the US or by latitude/longitude. We have captured the diversity of those results below.

Fourth Question: Does the service deliver as promised?

Probably the most important question, before any buying decision is completely opaque today. Therefore, Agents Trust probes every listed endpoint and reads what settled on chain. Payment is not delivery: we do not yet verify delivery or score the quality of results — that is the next layer we are building.

Back to the weather forecast API example: Seller Greeneris listed a coordinate forecast at $0.001 USDC on Base, the lowest price. We could attribute no earlier settlement to the advertised seller wallet. It answered our request, and we paid settlement on Base. The response body contained a 48-hour hourly series plus two daily rows for Zurich, with units on every field. In this test, it matched Open-Meteo within 0.2 mm of rain.

So, low usage is not a sign of pure quality necessarily. We have found that this situation generalizes from "weather forecast" to other services.

The x402 ecosystem is early stage and is in desperate need for reliable delivery quality attestations.

Conclusion

Catalog membership is not evidence of availability, payment history is not evidence of quality and metadata is not evidence of delivery. An autonomous buyer needs all four: availability, usage, relevance and delivery.

Agents Trust provides that evidence layer by combining catalog discovery, liveness probing and settlement history; observed delivery is the layer we are adding next.