Guides · Use our data

How do I use this site’s data before my agent spends?

Before your agent pays a seller, three pages here answer three questions with public evidence: who earns (rankings), what is for sale (directory), and how much money moves (market). This guide shows how to read each one — and how your agent can read the same data itself.

How do I read the rankings?

The rankings page orders sellers by one measure: verified revenue over the last 30 days. Not claimed revenue. Not popularity. Money we observed arriving.

Verified has a narrow meaning here. It is USDC that actually landed at the seller’s address on the Base network, read from the public ledger, with exchange deposits, fan-out senders, and self-transfers excluded by a published method. If we could not recompute a payment from the chain, it is not in the number.

Why rank by that? Because it is the one figure a seller cannot claim into existence. The 30-day window also slides — a seller keeps its rank only by keeping buyers. Beside revenue, each row shows how many distinct buyers paid in the window, and how much of the revenue came from the single largest one. Most revenue from one buyer is concentration risk — worth understanding before you depend on that seller.

Some rows carry an amber farm patterns badge. It means the seller shares infrastructure patterns with suspected listing farms — for example, one address standing behind a very large block of listings. It is a signal to look closer, not an accusation. Flagged sellers stay listed, never hidden, and the detection method is published.

How do I read the directory?

The directory marks every listing from on-chain evidence, and each row carries one of two markings. ✓ Verified revenue means we observed real, recomputable payments to that seller. Listed only means the service appears in a catalog, but we have not observed verified revenue for it.

Listed only is not a demerit. The service may be new, low-volume, or paid over rails our Base lens cannot see. The difference between the two markings is evidence, not quality — and absence of evidence is stated as exactly that.

The price column is the median on-chain payment to the seller over 30 days — what buyers actually paid, not what the catalog says the service costs. Catalog-asserted prices are claims we cannot verify, so listed-only rows show no price at all. When a seller’s claimed price and its on-chain median tell different stories, ask why before your agent pays.

How do I read the market page?

The market page shows daily volume as two series, side by side. Verified volume passed the same exclusions as the rankings. Dust is the swarm of near-zero payments — probes, tests, spam. Dust can dominate a day’s transaction count while adding almost nothing in dollars, so it is always displayed beside verified volume and never netted out.

One scope line matters more than any figure on it: everything there is tracked-catalog volume — payments to the seller addresses our catalog tracks. It is not “the x402 market,” and it is not the whole agent economy. The tracked catalog is growing, and the page says plainly how many days of coverage it can show.

How does my agent read the same data?

Over a free JSON API — no key, no account. The entry point is /api/v1/meta: it names the as-of timestamp, the method version, the update cadence, and the current catalog snapshot. Teach your agent to check freshness there before trusting anything downstream — a number without its as-of date is worth little.

The full route list is machine-readable in the OpenAPI description at /api/v1/openapi.json. Both endpoints answer as of 2026-09-02. One parsing convention to know: money fields are integers in micro-USDC — millionths of a dollar — so nothing is lost to rounding.

What is the cite button on every figure?

Hover over any published number on our data pages — on a touch screen, tap it — and a small “cite” button appears. Pressing it copies a plain-text citation: the figure, the page and section it lives in, its as-of date, the method version, and the recompute path. Quote us with the trail attached, so anyone reading your quote can check it the same way you can.

Should I trust these numbers?

No. The standing rule of this site applies to this site: never trust our numbers either — recompute them. Every published figure is recomputable from public inputs, and we publish all of them:

• The method — step by step, in the methodology.
• The exclusion lists — which addresses were flagged as exchanges, fan-out senders, or self-transfers, each entry with its source, at /data/flags/{version}.
• The raw catalog snapshots — the untouched catalog data a computation started from, at /snapshots/{id}.

You do not have to guess the current version or id. Every seller profile in the API carries a provenance block naming the exact flags URL and snapshot URL its numbers were computed from, and /api/v1/meta names the current catalog snapshot. If your recompute disagrees with ours, report it — the contact path is on About. A confirmed correction improves the data for everyone.

What does OTTO check before it buys?

OTTO, our example agent, is an invoicing agent that buys OCR — text extraction from scanned pages — at $0.01 per page. Before OTTO’s operator lets it spend, five checks, in order:

1. Directory. Does the OCR service carry ✓ Verified revenue, or is it listed only? Listed only does not disqualify it — it means the remaining checks lean on the seller’s own evidence instead of ours.
2. Rankings. Does the seller’s row show buyers beyond a single large one? A top-buyer share near the whole deserves an explanation before OTTO joins.
3. Price sanity. Is what buyers actually pay consistent with a service priced at $0.01 per page? A cents-priced service whose median payment tells a very different story deserves a question first.
4. Freshness, agent-side. OTTO fetches /api/v1/meta to confirm the data is current, then the seller’s API profile — with its provenance block — before any 402 flow begins.
5. Cap it anyway. A spending cap bounds whatever the checks missed.

These are the same signals as the manual checklist in How to evaluate a seller — precomputed. When our data and your own reading of the chain disagree, the chain wins. That page shows how to read it directly.