What Is Actually Priced: The One-Cent Ceiling, and the Market That Escaped It

Two markets pay for the same underlying work — find something out, check it, hand it over. One pays one cent. The other pays thirty-five to eight hundred dollars. The gap is not technology, and this page argues it is not even value. It is that one of the two markets can tell whether the work was any good.

What agents actually pay, counted twice

Salesmart S.r.l., which publishes Sinapsi, enumerated the whole x402 discovery index on 3 August and again on 5 August 2026 (The x402 Market, Counted: 14,766 Listings, 12,741 Calls a Day, One Winner). Prices, across 28,191 declared offers:

| | | |---|---| | Median price per call | $0.010 | | Interquartile range | $0.004 – $0.050 | | Under ten cents | 81% of offers | | One dollar or more | 2.4% |

That is the ceiling. Almost nothing in this market is allowed to cost as much as a coffee.

The number that explains the ceiling: nobody comes back

Price is what a market says. Repeat purchase is what it does. Every listing reports both its 30-day call count and its 30-day count of distinct payers, so the ratio is computable for all 14,457 listings that had at least one of each on 5 August.

| Calls per distinct payer, 30 days | | |---|---| | 25th percentile | 1.00 | | Median | 1.00 | | 75th percentile | 2.00 | | 95th percentile | 9.0 | | Listings where the average buyer calls fewer than 1.5 times | 8,391 — 58% |

The median paying customer of the median paid resource buys once and never returns.

It gets sharper when you look at the listings with the *most* customers. These are the eight most widely-bought resources in the index on 5 August, and they are the ones with the least return:

| Distinct payers | Calls | Per payer | Resource | |---|---|---|---| | 587 | 928 | 1.58 | api.onesource.io/api/chain/block-number | | 554 | 571 | 1.03 | api.onesource.io/api/chain/ens/:input | | 552 | 554 | 1.00 | api.onesource.io/api/chain/erc20-balance | | 519 | 522 | 1.01 | api.onesource.io/api/chain/live-balance | | 458 | 461 | 1.01 | api.onesource.io/api/chain/tx/:hash | | 450 | 460 | 1.02 | api.onesource.io/api/chain/block/:number | | 450 | 454 | 1.01 | api.onesource.io/api/chain/total-supply | | 448 | 454 | 1.01 | api.onesource.io/api/chain/contract/:address |

Five hundred and fifty-two agents paid for an ERC-20 balance, and five hundred and fifty-four calls were made. That is not a customer base. That is five hundred and fifty-two tastings.

There is a trap here worth naming, because we walked into it. Comparing the two snapshots, the distinct-payer count on these resources rises — 558 → 587 on the first one over two days — and that looks like the single healthy signal in an otherwise flat market. It is not. Calls rose in lockstep, roughly one per new payer, so what grew was the number of agents trying it once. A distinct-payer count with no repeat behind it measures curiosity, not demand, and it is the metric most likely to be quoted as evidence of growth.

Only one shape shows real return, and it is always the same one — retrieval. Selection rule, stated so it can be checked: highest calls-per-payer among listings with at least 40 distinct payers. The floor matters — without it the top of the ranking is occupied by single-payer endpoints such as agents.chain.link (26,263 calls, one payer), which are one integration billing itself rather than a market (The x402 Market, Counted: 14,766 Listings, 12,741 Calls a Day, One Winner):

| Payers | Calls | Per payer | Resource | |---|---|---|---| | 47 | 113,279 | 2,410 | x402.twit.sh/tweets/search | | 421 | 58,352 | 139 | x402.tavily.com/search | | 90 | 3,863 | 43 | api.exa.ai/search |

A handful of integrations that found something they genuinely need, and everything else being sampled once. Note also that the index's own headline is falling: 382,225 calls on 3 August, 378,427 on 5 August, with 470 listings appearing and 616 disappearing in forty-eight hours — 7.4% of the whole index churned in two days.

The market next door, paying four orders of magnitude more

Lead generation is an agent economy that predates x402 by decades: somebody finds a prospective customer, hands the record over, and gets paid per record. Published prices (quoted verbatim, self-reported by the industry):

> "Personal injury commands $200-$800+." > "Auto insurance $15-$75, home $20-$100, life $25-$125, health $30-$150, Medicare $30-$100." > "Exclusive qualified leads $100-$200+, shared $35-$125, set appointments $150-$200, aged > (30+ days) $5-$30."

Against a $0.010 median per agent-market call, the cheapest shared lead quoted above — $35 — is 3,500×, and an $800 personal-injury lead is 80,000×.

Why they are allowed to charge that

Three mechanisms, none of which exists in x402:

Inspection before payment. In *ping-post*, quoting the mechanism verbatim, the seller *"transmit[s] partial lead information — known as the 'ping' — to multiple potential buyers simultaneously"*; *"once buyers evaluate the ping, they respond with a bid (or a rejection)"*; and the system then *"routes the complete lead details — the 'post' — to the winning buyer"*, all *"in real-time, typically within a matter of seconds"*. The buyer commits money against a description and takes delivery against the thing itself. That is structurally what a paywalled cover over withheld content does, and what the 402 flow does *not* do: in x402 the money moves first and the content arrives afterwards, with nothing in between.

Reversal after inspection. The industry's own definition:

> "Return rates represent the percentage of delivered leads that buyers reject after purchase > and successfully claim credits for."

And it is not marginal. Healthy target bands run from 4-6% in legal to 12-15% in solar, with each vertical declaring the level at which a source becomes a problem. A market that budgets for between one lead in twenty (legal) and one in eight (solar) coming back is a market where payment is provisional until the buyer has judged the goods. x402 has no such state. Once settled, settled.

Reputation attached to the source, not the seller. Buyers track returns per source and throttle the bad ones. Compare what the agent market can see: its quality object counts calls and payers — demand — and nothing anywhere records whether a response was correct (The x402 Market, Counted: 14,766 Listings, 12,741 Calls a Day, One Winner). The trust services that grew around the index measure liveness, uptime and wallet plausibility, never outcome, because they sit outside the transaction (The Agent Labour Market, Probed: A Complete Bounty Board With Four Customers).

Nobody else is building the missing layer either

The gap is not only a market failure waiting for a startup. It is also absent from every rulebook written so far, and the regional pages of this wiki record the same shape three times over:

- India has gone furthest — a payments regulator, the national payments operator and a cyber agency have all put agent spending in writing — and every instrument is a cap on *quantity*: amount, frequency, validity. None constrains conduct (India Is Writing the First Agent-Payment Rulebook, and Every Line of It Is a Number). - Brazil has the rail an agent would design for itself — instant, account-to-account, API-addressable, with a native recurring mandate since June 2025 — and no agent protocol running on it (Brazil Built the Best Non-Crypto Agent Rail and Put No Agent Protocol On It). - Korea and Japan shipped the visible half: consumer shopping agents inside dominant platforms, spending from wallets that already existed, with no agent payment rail underneath and no published agent-initiated transaction count (Korea and Japan: Shopping Agents Shipped, Agent Payment Rails Did Not).

Three jurisdictions, three different starting points, and the same missing piece as x402 — which is evidence that the omission is structural rather than an oversight by one protocol's authors (The EU Wrote the Rules First and Left the Agent Out of Them, US and Canada: The Rails Shipped First, and the Rulebook Named the Payer).

The reading

This part is our interpretation, not a measurement. The one-cent ceiling is not a statement about what information is worth to an agent. It is the price of a lottery ticket. A buyer who cannot tell a good seller from a bad one before paying, cannot get the money back after, and cannot consult anyone's experience of the outcome, can rationally risk about a cent. Raise the price and the expected loss from the unknown swamps the expected gain.

That is why the demand curve looks the way it does: 58% of listings sampled once and abandoned is exactly what a market of one-cent lottery tickets should look like. Buyers are not using these services. They are finding out, one cent at a time, that they cannot tell.

The lead market escaped the same trap not by inventing better payments — it settles on ordinary invoices — but by building the layer x402 skipped: inspect, reverse, and remember who was right (A Spending Limit Is Not a Conduct Limit).

This is also the practical shape of the problem for anyone selling to agents: the scarce good is not payment, which works fine, and not content, which is abundant. It is evidence that a particular source held up (The Earn-Spend Loop: Why Machine Payment Is Half-Built, The Binding Constraint Is Verified Context, Not the Model or the Rail (our prediction)). A seller who can produce that evidence is not competing in the one-cent market at all — they are in the market next door (Four Ways to Sell Content to an AI, and Who Keeps What).

Falsifier

We are wrong if, by 30 June 2027, the median calls-per-payer across the discovery index rises above 3.0, or the median listed price rises above $0.05, without a conduct layer (inspection, reversal or outcome-linked reputation) having appeared. Either would show the ceiling was a stage of adoption rather than a consequence of missing information. Both are recomputable from a fresh snapshot with the script in agent-economy/dati/ — one command, and it contradicts us or it does not.

We are also wrong if a repeat snapshot shows the one-call-per-payer pattern is an artifact of how the operator counts distinct payers (for example, a new wallet per call). We have not been able to rule this out from outside, and it is the single assumption on which the central finding rests.