Executive Summary
Status code 402 was reserved for a payment challenge that ordinary web stacks rarely honored. The specification at x402.org binds that status to a closed client-server loop: an unpaid call is answered with a challenge, the caller transfers stablecoins, and a retry that carries proof receives the resource. The same page states that the loop rides inside ordinary HTTP, that a server opts in with one middleware registration, and that production use has already reached millions of transactions. A separate product page, Coinbase's developer overview, reports that its own facilitator has handled more than 100 million payments on Base and Solana. Those sentences measure different things. One is a protocol-level production claim. The other is a single vendor's cumulative count on two named networks. This essay keeps the two figures apart, reconstructs the request loop from the pages that state it, and writes the economic swap from prepaid accounts to per-call settlement in symbols. No tariff in the fetched pages is numeric, so none is invented here.
The falsifiable mechanism is narrow. If a server can demand value and release a resource inside one HTTP conversation, opening an account is no longer a prerequisite for a machine client. The project homepage lists five steps for the older pattern and three steps for the replacement. Everything after that list is either a direct quotation of scope or an explicit inference.
What the status code does
The homepage describes the server side as route-scoped middleware. The worked example attaches a handler to GET /weather. Configuration includes an accepts array — the comment in the snippet says "as many networks / schemes as you want to support" — and a short description of the resource. An unpaid arrival is answered with HTTP 402. The client pays and retries. The resource follows. The page's constraint on topology is one sentence: the exchange is built into existing HTTP requests, with no additional communication required. Challenge and proof therefore live in that request/response cycle. The fetched text never names a header, a body field, or a signature algorithm. This essay does not supply the missing names.
The Coinbase overview describes the buyer side of the same loop. The caller sees a price, pays, and gains access without leaving the request flow. The overview states that there is no checkout page, no card form, and no signup flow. It names three buyer shapes: a person unlocking premium material, a backend purchasing metered data, and an agent buying a tool call while a task is still running.
| Order | Older access path, as listed | Replacement path, as listed |
|---|---|---|
| 1 | Create an account with the provider | Caller sends the HTTP request and receives 402 |
| 2 | Add a payment method; identity checks delay access | Caller pays at once in stablecoins; no signup and no approval |
| 3 | Purchase credits or a subscription | The resource is granted |
| 4 | Store and rotate an API key | — |
| 5 | Pay, and accept slow transfers, chargebacks, and fees | — |
The empty cells are empty because the second list ends at three items. The fifth older item is the only place chargebacks appear.
The FAQ on the project homepage states the intended jobs: instant, low-cost payment for digital services, API monetization, agentic commerce, paywalled content, and any setting where older methods are called too slow or too expensive. "Low-cost" is the page's adjective. No schedule of prices sits beside it. Chain support is stated as blockchain-agnostic: all EVM-compatible chains, Solana, and more. Stablecoin settlement is the primary use case. The same answer says the design is extensible to traditional payment methods. The copyright line places the effort under a Series of LF Projects, LLC. Read as an operator, the practical consequence of the three-step list is that a new machine client can reach a priced endpoint before any credential exists on the seller's side. That property is what the account-based list spends its first four steps constructing.
Hosted verification against a neutral spec
The homepage compresses its design claims into five negatives. A specification-layer surcharge is absent: customer and merchant "just pay nominal payment network fees." Delay is rejected by slogan: "money moves at the speed of the internet." Accounts and personal information are not required. Permission to implement is open: "anyone on the internet can build on or extend" the work. A home chain is disclaimed in the phrase "a neutral standard, not tied to any specific network." The FAQ adds that the implementation is open-source and has been audited for security, and that it is production-ready.
Coinbase's overview describes a service that sits on that specification. A builder can ship a paid endpoint without constructing chain infrastructure. The CDP Facilitator verifies and settles. Sellers are paid at any address they control. Four destination classes are named: a CDP custodial wallet, Coinbase Business, Prime, and a self-custody wallet. A buyer pays from a CDP wallet or from another signer. The overview points onward to a page titled "How x402 works" for the loop in one place, and to facilitator notes covering networks, pricing, and production readiness. Those later pages were not in the source set. Their prices, chain menus, and readiness checklists are therefore not restated here.
The tension between the two documents is an inference, marked as such. Neutrality answers who may implement. The hosted verifier answers who will check a proof if the seller does not. A caller that can speak only to one settler has an operational dependency the homepage does not require. A seller that chooses self-custody still relies on the verifier whenever verification is outsourced. Hosting is a product. It is not, on the text of the homepage, a rule of the standard.
sequenceDiagram participant Caller as Client or agent participant Origin as Resource server participant Verify as CDP Facilitator participant Ledger as Base or Solana Caller->>Origin: Request the resource Origin-->>Caller: HTTP 402 and accepts list Caller->>Ledger: Stablecoin transfer Caller->>Origin: Retry with proof Origin->>Verify: Verify and settle Verify->>Ledger: Settlement submission Origin-->>Caller: Resource body
The drawing follows the homepage loop, then adds the verifier the overview actually names, on the two networks that overview counts. A deployment that checks proofs inside the origin matches the homepage wording and simply omits the middle participant. Finality depth, retry deadlines, and exact payload layout are unspecified in the fetched paragraphs, so they are unspecified here.
Base is one of the two networks in the vendor tally. The Base documentation index was fetched with this brief and does not print a payment total for this scheme. The Base docs path likewise contributes no such total in the material reviewed. The cumulative count stays where it was printed: on the facilitator overview, not on the chain's documentation shell.
Expected cost, in symbols only
Neither fetched page states a gas price, a median fee, or a subscription tariff. A comparison that inserted cents would be fiction. The model below keeps every money term as a variable. The sources decide only which variables exist.
Expected cost of one paid call on the new loop:
is the price carried in the challenge. is the nominal network fee that remains after the specification-layer surcharge is removed. is the chance the first proof fails and the caller must try again. is the extra cost of that second attempt. No source estimates .
Expected cost of one call on the five-step list, spread across later calls:
is account creation, which the homepage calls time-consuming. is access delayed until approval. is the burden of storing and rotating a credential, which that list calls a security risk. is the credit purchase or subscription. is balance left unused when the buyer overpays. The same bullet also names the opposite failure, running out of funds; that stockout is a separate loss and is not given a rate here. and stand for the fees and chargebacks named in the fifth older step.
A machine client prefers the new loop when:
For a fixed onboarding bill, a short relationship (small ) favors the 402 path because setup, approval, and key custody are paid once under the old list and then diluted. A long relationship can favor a subscription if is charged on every call and the server's stays above the marginal cost of serving the body. That crossover is a shape. It is not a measured threshold. The homepage says the new path does in moments what older systems cannot do at all. Read against the lists, that sentence is about , , and dropping out, while is explicitly retained.
Irreversibility is the paired risk, again as inference. Chargebacks occur only on the older list. The replacement list has no reversal step. Dropping chargebacks drops a buyer remedy and drops an issuer cost. An agent that cannot file a dispute never held that remedy. A person buying paywalled content did. The overview's own split — human, backend, and agent — therefore does not share one loss function. The homepage's "zero friction" claim and the disappearance of chargebacks are the same design choice seen from opposite sides.
Adoption, limited to printed receipts
Two quantitative statements are on the pages, and they do not match in scope.
The FAQ says the system is production-ready and has processed millions of transactions, and that the code is open-source and audited. "Millions" fixes an order of magnitude. It is not a ledger extract, a date range, or a chain split.
The facilitator overview says CDP has processed more than 100 million payments across Base and Solana. That sentence binds a vendor, a payment definition, and exactly two networks. It can sit inside a loose reading of "millions," or it can dwarf that word. The pages never reconcile the pair. Calling the vendor figure the scheme's global total would add a scope the FAQ does not state. Calling "millions" a rebuttal of the vendor figure would erase a sentence that is printed. The conjunction is the whole result: the homepage asserts a multi-million production history, and one named settler asserts a nine-figure cumulative on two chains.
Non-numeric signals on the homepage include a member list, working groups, a meeting calendar, and a community described as thousands of builders on an open codebase. None of those is a settled transfer. The overview adds distribution doors: a seller SDK, a buyer SDK, agentic accounts (an account the operator hands an agent so the agent can pay), and integrations toward platforms where agents already run. Doors are not volumes.
The following are absent from the fetched text and are not asserted: a split of payers between humans and agents, a distribution of amounts, a failure rate, a median time-to-inclusion, and any fee numeral. "Zero wait" is a homepage slogan. Settlement on the two named networks still has whatever finality those networks have. The Base documentation shell reviewed here does not state a finality parameter, so none is filled in. The overview says the facilitator verifies and settles. It does not, in the fetched paragraphs, define whether the origin may return the body before inclusion. That predicate is the open measurement.
Source-to-claim map
| Statement used above | Printed at | Not imported |
|---|---|---|
402 challenge, retry, one middleware registration, accepts array, no extra channel |
Project homepage | Header names, signature schemes |
| Five older steps, three replacement steps, no key custody on the new list | Same homepage | Minutes of onboarding |
| No specification-layer surcharge; nominal network fees remain; no home network | Same homepage | A fee integer |
| Millions of transactions; open-source; audited; EVM chains plus Solana and more; stablecoins primary; traditional methods called extensible | Same homepage | Auditor name, chain census |
| No checkout, card form, or signup; person, backend, and agent as buyers | CDP overview | A mix ratio |
| Hosted verifier settles; four payee classes; payer is a CDP wallet or another signer; vendor cumulative on two networks | Same overview | Gas, latency, a third chain's share |
| Chain documentation shell lacks the payment total | Base index and Base docs path, linked above | Any throughput figure |
Predictions
Each line is a check against a public page on a date. The confidence figure is the author's score for that check, not a frequency measured in the sources. A miss on the date falsifies the line.
- By 30 September 2027, a fresh fetch of the Coinbase developer overview will print a cumulative payment count, for the two networks it already names, numerically greater than the count in the fetch used for this essay. Author confidence: 58%.
- By 31 December 2027, the project homepage will show a concrete production path for a traditional, non-stablecoin payment method, matching the extensibility sentence already published there. Author confidence: 36%.
- By 30 June 2027, either that homepage or the Coinbase overview will name a production verifier other than the CDP Facilitator as settling live calls. Author confidence: 44%.