We ran Coinbase's own validator against every endpoint in Coinbase's own x402 catalogue — all 10,811 of them. Just over half fail it. Nearly a third cannot accept a payment at all.
Measured 18–19 September 2026. Rechecked continuously, oldest first.
Every resource listed in the CDP Bazaar discovery catalogue, submitted to the CDP validator endpoint, which applies 25 preflight checks. The verdicts are not ours. They are the catalogue's own validator applied to the catalogue's own listings, so no endpoint here is being judged against an outside standard.
| Result | Endpoints | Share |
|---|---|---|
| Pass all checks | 5,179 | 47.9% |
| Fail one or more | 5,632 | 52.1% |
Lumping them together hides the important part. A listing that is merely hard to discover is a different problem from a listing that takes your money nowhere.
| Failure mode | Endpoints | Share of catalogue |
|---|---|---|
Broken. No parseable 402 at all — wrong status, unparseable body, or nothing there | 3,276 | 30.3% |
| Invisible. Payment works, but discovery metadata is missing or malformed | 2,294 | 21.2% |
The first group is the one to care about as a buyer. Close to a third of everything in the catalogue is listed, looks live, and will not complete a payment. We spot-checked samples directly and got exactly what the validator got: 404, 405, 502, and a handful of endpoints returning 200 with no payment step at all.
The second group is a seller's problem, and a quiet one — the endpoint works perfectly. It just cannot be found, priced, or called automatically by anything that reads the catalogue.
Almost everything listed has been paid at least once: 10,807 of 10,811 endpoints have one or more unique payers in 30 days. That number flatters the market. The distribution behind it does not.
| Unique payers in 30 days | Endpoints | Share |
|---|---|---|
| At least 1 | 10,807 | 99.96% |
| 10 or more | 317 | 2.93% |
The median endpoint in the catalogue has one unique payer and one to two calls across the whole month. The mean of 3.97 payers is carried by a thin tail — a single endpoint tops out at 2,255. Read the median, not the average.
This is the finding we most expected to be an artifact, so we tested it that way. Mean payers alone would be misleading, because a handful of very busy endpoints distort both groups. After discarding the top 1% of each group, the gap survives:
| Passing | Failing | |
|---|---|---|
| Mean payers, top 1% removed | 2.33 | 1.64 |
| 90th percentile payers | 5 | 3 |
| 99th percentile payers | 26 | 17 |
| Median calls | 2 | 1 |
Every robust statistic points the same way, by roughly 40%. But the medians are equal at one payer each. Conformance is worth having and it is not what decides whether anyone shows up.
Check it before you call it. A 30.3% chance the thing you found in the catalogue cannot complete a payment is worth one cent to rule out.
Discovery metadata is the single most common thing missing: 2,294 endpoints that take payments correctly are invisible to anything reading the catalogue. Among failing endpoints, the most frequent individual failure is a missing output example, at 67.8%.