Why would a buying agent skip an endpoint that correctly returns 402?

Most often because the 402 names no facilitator. A spec-following client reads the facilitator from accepts[0].extra.facilitator, then accepts[0].facilitator, then the document root. Named in none of the three, the client falls back to a default that cannot settle to the seller, so a healthy endpoint reads to the buyer as a payment failure. Other common causes are a Payment-Required header grown past a reverse proxy's default 4KB buffer, incomplete terms that cannot be signed, and a facilitator URL that no longer answers.

Does the x402 Endpoint Report need access to my server or logs?

No. It calls the endpoint from outside with no payment and no credentials, exactly as an unknown buyer would, and reads the 402 response. Nothing is installed and no traffic data is shared, which also means it reports the buyer's experience rather than the seller's conversion rate.

Can it tell me how many agents tried to pay and gave up?

No, and any tool claiming otherwise from an outside probe is guessing. True conversion requires the seller's own request logs and settlement records. The report measures whether a payment is possible and economic, which is the part a seller cannot see from the inside.

What does the score mean?

It starts at 100 and subtracts 25 for each high-severity defect, 10 for medium and 4 for low. 90 and above reads as buyer ready, 60 to 89 as payable with defects, 30 to 59 as buyers will struggle, and below 30 as effectively unbuyable. A perfect score means nothing testable is broken, not that agents will certainly buy.

Why does the report flag a price as too low?

Because settling a payment costs gas. Measured from real transferWithAuthorization receipts, a settlement uses about 86,262 gas units rather than the 65,000 commonly quoted. When that cost exceeds the price of the good, a buyer that broadcasts its own payment loses money by trading with you, so the rational agent declines. Routing through a facilitator that absorbs the gas removes that floor.

Is the check free, and is there a paid version?

A single endpoint check is free with no signup. Batch reports across every endpoint a seller runs are live at $0.01 per call. Scheduled re-probing that alerts when a passing endpoint regresses is in private beta and available by request.

Your paid endpoint answers 402. Agents still are not paying. Here is what they see.

Paste any public x402 endpoint. We call it from outside with no payment and no credentials, then read the 402 back to you the way a buying agent reads it.

Free, one endpoint per check, no signup. We never send a payment and never see your traffic.

What it checks, and why each one loses you money

Check	What goes wrong when it fails
Facilitator named at all three levels	A spec-following client reads the facilitator from accepts[0].extra, then accepts[0], then the root. Named nowhere, it falls back to a default that cannot settle to you, so a healthy endpoint reads to the buyer as a payment failure.
The facilitator actually answers	An advertised facilitator that 404s on /verify is worse than none: the buyer trusts your terms and stalls. We probe the one you name.
Payment-Required header present	Coinbase's indexer reads the terms from that header, not the body. Absent, the endpoint cannot be listed at all, however perfect the JSON is.
Header survives a default proxy	Past 4KB, nginx's default buffer drops it and the buyer gets 502 while your app logs a clean 402. Silent, and it looks like their bug.
Terms complete enough to sign	Missing payTo, asset, network, amount or maxTimeoutSeconds and the agent cannot construct a payment even when it wants to.
Price clears the settlement floor	Priced below the cost of broadcasting the payment, a buyer that settles for itself loses money by trading with you. We measure that floor from real settlement receipts, not a textbook estimate.
Discoverable and describable	No serviceName, description or schemas, and an agent must guess what it is buying. Most do not guess.

Scope, stated plainly

This is probed from the outside. It describes the buyer's experience, not your traffic, and needs nothing installed.
It cannot tell you how many agents tried, because we do not see your logs. If you want true conversion, that needs your data and is a different conversation.
A score of 100 means nothing we know how to test is broken. It is not a guarantee that agents will buy.

Beyond a single check

Batch — every paid endpoint you run, in one report, ranked by what is costing you the most.
Monitoring — re-probed on a schedule with an alert when a passing endpoint regresses. Specs rot quietly: ours advertised a dead facilitator for months before a buyer told us.
Batch is live. Scheduled monitoring is in private beta. Mail [email protected] with your endpoint list and we will run it by hand first, free.

Batch audit — live

Audit up to 20 endpoints in one call for $0.01 USDC on Base. Same checks as the single report, plus a defect distribution across the set. Endpoints we could not reach are returned as cannot judge rather than scored, because a blocked probe is not a finding.

POST https://x402-agent-pay.com/facilitator/heatmap/batch
{"urls": ["https://example.com/api/paid", "..."], "method": "GET"}

We ran this check across every endpoint listed in the public x402 directories: read the findings — 13 of 282 are still payable, and all 13 omit the facilitator field.