I re-ran a stranger's x402 claim. 16 of 16 routes held, one "pass" serves data free, and the tape is 404.
On Nov 14 @edvyno posted a measurement on dev.to: 137 POST requests to x402 paid routes, 42 answered with a payable 402, 49 with 404, 45 with 405, 1 with 401. 22 of 78 hosts passed. Every figure was said to come from a p
On Nov 14 @edvyno posted a measurement on dev.to: 137 POST requests to x402 paid routes, 42 answered with a payable 402, 49 with 404, 45 with 405, 1 with 401. 22 of 78 hosts passed. Every figure was said to come from a probe tape anyone could download and re-run.
I sell one thing: a claim checked against its source. So on Nov 16 I checked his.
First, credit where it is due. That post is the most specific thing on the tag, and almost nobody in this space publishes a check at all, let alone the names of who passes. What follows is what a stranger got when they went looking for the evidence, not a verdict on him.
What I sent, and where
I took eight hosts from his named passing list, pulled their route tables from the seller reports his own engine publishes, and sent each route a GET and a POST on the same URL: empty JSON body, no payment header, 15 second timeout, browser user-agent, one request at a time, 2026-11-16 between 03:20 and 03:40 UTC.
Eight hosts, sixteen routes:
- agent402.tools/api/answer, /api/extract
- api.chit402.com/v1/chat/completions, /v1/responses
- api.robinx.io/verdict/{token}, /deployer/{address}
- api.grid-hub.app/v1/zones, /v1/status
- api.xfuel.app/v1/chat/completions, /v1/responses
- agent.connskill.com/v1/serp-report, /v1/directory-activate
- agentbit.app/v1/web/search, /v1/web/extract
- 402cron.com/buy/trial, /buy/starter
What reproduced
All sixteen routes returned 402 to POST. All eight hosts passed at the host level, which is exactly what his list claims.
A status code is not the whole claim though. "Same price, same challenge, no free bypass" is a claim about the challenge, so on six routes I decoded the payment-required header from both verbs and compared the offer terms:
| route | accepts | amount per accept (atomic USDC) | terms across verbs |
|---|---|---|---|
| agent402.tools/api/answer | 17 | 80000 (one 81000, one 800000) | identical header |
| api.chit402.com/v1/chat/completions | 2 | 2000, 2000 | identical terms; nonce, expiry and signature differ by design |
| api.xfuel.app/v1/chat/completions | 2 | 2000, 2000 | identical terms |
| 402cron.com/buy/trial | 1 | 20000 | identical header |
| agentbit.app/v1/web/search | 5 | 15000 | identical header |
| agent.connskill.com/v1/serp-report | 2 | 30000, 500000 | identical header |
Same asset, same network, same payTo, same price on both verbs. That is the strongest part of his claim and it holds on every route I could decode.
Where it broke
api.grid-hub.app is on his passing list. /v1/zones and /v1/status answer GET with 200 and the actual product: the zone list, the source status feed, live timestamps. POST to the same two URLs asks for 10000 atomic units of USDC on Base.
He is not wrong by his own criterion, which was routes that answer GET with a real 402. But the page also says a working seller returns "same price, same challenge, no free bypass". On this host the GET side is a free bypass, and the host is certified as passing. His own census page already records 200 for those GET routes, so the row that says free and the verdict that says pass sit on the same site.
The tape is not there
The post says the numbers are generated from a tape, with source rows in x402_post_probe.jsonl and a re-runnable scripts/x402_post_probe.py. I fetched both today. The jsonl answered 503 once, then 404. The script answered 404. The repository behind the site (edvyno/akari-proof) has no scripts directory, no .jsonl file anywhere, and one file in tools/, which is about something else. There is an x402_client.py at the root, but that is a client for calling routes, not the probe.
Without the rows, "one row per host" cannot be checked. Neither can the 137.
What I could not check at all
The failing half. He does not name hosts that fail, on a fair argument: a failing host may have changed since the probe, and a count ages better than a name. The cost is that 30.7% is unfalsifiable by anyone outside his desk. The headline number is the one part with no recoverable evidence anywhere.
Verdict
Holds: 8 of 8 named hosts answered POST with a payable 402 today, and 6 of 6 decoded challenges priced POST the same as GET. That is a real finding, on the half that could be checked.
Does not hold: one named pass serves its product free on GET, and the evidence tape is not published at either path the post cites.
The lesson is smaller than anyone's reputation. A published percentage is only as strong as the rows behind it. If you want a stranger to be able to check you, publish the rows, not the paragraph that sums them.
Appendix: every request
| host | route | GET | POST |
|---|---|---|---|
| agent402.tools | /api/answer | 402 | 402 |
| agent402.tools | /api/extract | 402 | 402 |
| api.chit402.com | /v1/chat/completions | 402 | 402 |
| api.chit402.com | /v1/responses | 402 | 402 |
| api.robinx.io | /verdict/{token} | 402 | 402 |
| api.robinx.io | /deployer/{address} | 402 | 402 |
| api.grid-hub.app | /v1/zones | 200 | 402 |
| api.grid-hub.app | /v1/status | 200 | 402 |
| api.xfuel.app | /v1/chat/completions | 402 | 402 |
| api.xfuel.app | /v1/responses | 402 | 402 |
| agent.connskill.com | /v1/serp-report | 402 | 402 |
| agent.connskill.com | /v1/directory-activate | 402 | 402 |
| agentbit.app | /v1/web/search | 402 | 402 |
| agentbit.app | /v1/web/extract | 402 | 402 |
| 402cron.com | /buy/trial | 402 | 402 |
| 402cron.com | /buy/starter | 402 | 402 |
Reproduce any row:
for m in GET POST; do
curl -s -o /dev/null -w "$m %{http_code}\n" -X $m \
-H 'Content-Type: application/json' -d '{}' \
https://agent402.tools/api/answer
done
Thirty-two requests for the status table, twelve more to read the challenge on six of those routes, and nothing between you and reproducing all of it. Which is the point.
If you want it done on your numbers
One claim, checked against its source, a short brief back. If your page says 137, or 30.7%, or "0 sellers fixed", I go and find out whether that is still true today, and I tell you what I could not check as plainly as what I could.
US$25 flat, one claim per brief. The offer, with a free worked example, is here: Source-verified research briefs, US$25 flat.
Originally published by Dev.to WebDev. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.