The agent markets, measured

Cody Lee Walker · s2ar · 2026-10-10 · a living page: the numbers move with each daily snapshot

Every number on this page is a living number from the daily snapshot (the small ◆ beside it names the aggregate it came from, by hash) or carries a note to its source. The aggregates are in the site’s data/markets/agg/; the raw snapshots stay on the box that takes them. MEASURED — not argued.

The latest snapshot holds 31942 ▼2803 ◆ on 2201 ▲24 ◆; §5 says how many days the series has; the sources, the method and the limits are in §7.

1. The question

Agents now buy from other agents over HTTP, paying per call with a signed USDC transfer (x402). Two public indexes say what is on offer: Coinbase’s Bazaar, which lists a resource once a payment to it has settled through Coinbase’s facilitator, and Agent402, a router that crawls sellers’ manifests and decides whom to send buyers to. 11Coinbase, “x402 Bazaar discovery”, https://api.cdp.coinbase.com/platform/v2/x402/discovery/resources; Agent402, https://agent402.tools/llms.txt and /terms (programmatic use is accepted use). The question this page keeps asking, day after day: who actually gets paid, what it costs, whether anything in the market carries evidence of its own quality, and where s2ar sits.

2. Who actually gets paid

The Bazaar is not a catalogue of everything for sale. It is the set of resources that settled at least one payment recently: 31942 ▼2803 ◆ on 2201 ▲24 ◆ in the latest snapshot, almost all with at least one call in thirty days. ⊕The window is wide, not strict: the oldest lastUpdated in today’s snapshot is 2026-05-16, the newest today. A resource stays while something keeps paying it.

Resources by distinct payers in 30 days

Most listed resources have one or two paying wallets. The share with three or more distinct payers is 23.2% ▲2 ◆. 22Payer tiers from quality.l30DaysUniquePayers, aggregate 2026-10-10 (sha 9c5ac203…): tier 1 = 47.7 %, tier 2 = 31.1 %, tiers 3–9 = 20.0 %, 10+ = 1.1 %, zero = 11 resources. The median settled resource saw two calls and two payers in thirty days; the ninetieth percentile, five calls and three payers. 33calls_per_settled_item p50 2, p90 5, p99 74; payers_per_settled_item p50 2, p90 3, p99 11 (2026-10-10).

Who gets the calls: hosts vs calls

Calls concentrate hard. The top host alone takes 43.4% ▲0.3 ◆ of all calls in the window, the top ten 79.2% ▲0.2 ◆, and the Gini coefficient over hosts is 0.97. 44concentration: top-1 43.1 %, top-10 79.0 %, top-100 94.3 %, Gini 0.9731 over 2,176 hosts with calls; 934,193 calls in the window (2026-10-10). The top host is one resource with 4,692 payers; the second is a hundred resources with 237; the ninth is 11,255 resources with three. Listing many things and being paid by many wallets are different achievements, and the market rewards the second. ⊕Hosts stand in for sellers throughout: a host can be one wallet or several, and one wallet can run several hosts.

Agent402 adds the router’s view: of 4630 ▼51 ◆ it has indexed, 3447 ▲81 ◆ and 1371 ▼31 ◆ are ones it will actually dispatch a buyer to. The funnel narrows at every step.

Agent402: why a seller is or is not dispatchable

55Agent402 /api/index totals 2026-10-10: discovered 4,804, crawled 4,764, responded 3,947, routable 3,434, unhealthy 1,331, paid tools 137,820, distinct Base payees 2,194. Dispatch reasons: settlement_required 1,606, crawl_failed 1,314, eligible 800, settlement_checked_at_pay_time 602, network_unknown 247. The largest single reason a listed seller is not dispatchable is settlement_required: Agent402 waits for fifty settled payments from three distinct payers before it trusts a seller with its buyers’ money, a guard against sellers who pay themselves. A third of the rest failed the crawl.

Behind both indexes sits the chain. Agent402’s own series of gasless USDC settlements on Base, which counts every EIP-3009 transfer and not only x402, read 1195 ▼76058 ◆ from 343 ▼6081 ◆ on the last full day.

EIP-3009 settlements on Base per day (Agent402’s series)

66Agent402 /api/x402-economy (asOf 2026-10-09T23:11Z): last 7 days 780,563 settlements, 57,472 payers, 20,439 merchants, $51.99 M volume; last 30 days 2,604,073 settlements. The volume figure includes large non-x402 transfers; the settlement and payer counts are the usable numbers.

3. What it costs

Resources by price per call (USD)

The market prices by the cent. The median call costs 0.01 USD ±0 ◆, the ninetieth percentile 0.05 USD ±0 ◆, and 79.2% ▼1.5 ◆ of resources ask a cent or less. 77prices_usd over 34,635 dollar-priced resources: min $0.0002, p10 $0.002, p50 $0.01, p90 $0.05, p99 $0.75, max $1,000; 110 resources priced in a non-dollar token are left out.

Share of resources with ≥ 3 payers, by price band

Cheaper resources are paid by more wallets: at a tenth of a cent or less, 57 % of resources have three or more payers; between one and five cents, 11 %; above a dollar, 9 %. 88price_bands.share_ge3_payers (2026-10-10): ≤$0.001 0.571 (2,236 resources), $0.001–0.01 0.406 (8,493), $0.01–0.05 0.107 (20,410), $0.05–0.25 0.148 (2,730), $0.25–1 0.040 (575), >$1 0.093 (301). Price is not the only thing that differs between those bands, and the page does not claim it is the cause.

Resources by category (keyword rules, first match wins)

Categories are keyword rules over the listing’s name, tags, description and path, applied in the order shown in the method section, first match wins; they are crude and they are published. By those rules the largest groups are code, data and finance. 99categories (2026-10-10): code 10,205 · data 8,417 · finance 5,111 · evaluation 3,498 · other 2,652 · llm 2,372 · image 2,153 · infra 337; median price $0.01 in every group but llm ($0.006), infra and other ($0.005).

Share with ≥ 3 payers, by category

Finance and infrastructure listings are the most often paid by several wallets; code and evaluation listings the least. 1010share with ≥3 payers by category: infra 0.466, finance 0.372, other 0.281, data 0.253, image 0.213, llm 0.153, evaluation 0.125, code 0.114.

4. Where evidence sells

The house claim is that these markets lack what a buyer would need to trust a stranger: reputation, recourse and a verifiable history. The listings can be read for it. A listing “mentions evidence” when its name, tags or description contain a receipt-, certificate-, verification-, proof-, signature-, audit-, attestation-, replay- or determinism-word.

Evidence words in listings (resources matching)

5324 ▼644 ◆ resources on 803 ▲10 ◆ match, and 1,155 of them have three or more payers. Words are cheap: “deterministic” and “audit” lead the list, and a word in a description is a promise, not a mechanism. 1111evidence.by_word (2026-10-10): deterministic 1,571 · audit 1,379 · signed 600 · verified 575 · verification 513 · verify 431 · receipt 166 · proof 137 · certificate 118 · replay 77 · audits 70 · attested 46.

Mechanisms show up elsewhere, in the x402 extensions a listing declares. Beyond bazaar itself (every listing), the snapshot sees builder-code on 1,741, payment-identifier on 1,063, offer-receipt on 668, terms on 547, otto-content-receipt on 63 and sign-in-with-x on 76. 1212extensions_seen (2026-10-10). offer-receipt is the closest thing in the market to a receipt on the answer: a signed statement that a particular offer was served. None of the listings seen so far publishes a verifiable record of its own delivery rate, a signed daily account of calls paid and credited, or a charter it can be held to. That is the gap the Assay records and the proceedings are built for, and this page will say so only as long as the snapshot keeps saying so. ⊕The schema side is healthier than the evidence side: 99.98 % of listings carry an input schema, 90 % an output example, 48 % a service name, 78 % tags, 12 % an icon.

5. Dynamics

Snapshots so far: 2 (2026-10-10 → 2026-10-11). The window today holds 2201 ▲24 ◆ and 31942 ▼2803 ◆.

Since the previous snapshot: hosts new 62 ◆, hosts gone 26 ◆; the window changed by 31942 ▼2803 ◆ resources. Over all snapshots:

hosts 2201 · resources 31942 · share with ≥ 3 payers 23.2 · top host’s call share 43.4 · Agent402 eligible 1371 · Base settlements/day 1195

Resources in the Bazaar’s window, per snapshot

6. Where s2ar sits

s2ar’s seven paid routes entered the Bazaar on 2026-10-10 after the first settlements through Coinbase’s facilitator. In the window they carry 10 ▲3 ◆ from 1 ±0 ◆: not in the top hundred hosts, priced at the market’s median ($0.005–$0.02 a call), and, on Agent402, listed, healthy and routable but not yet dispatchable (settlement_required). 1313ours (2026-10-10): 7 Bazaar resources at 1 call / 1 payer each; Agent402 health 1, routable, 8 paid tools, routerDispatchEligible: false. That first payer was us, and the proceedings say so.

What a listing shows against what the proceedings show:

	a Bazaar listing	Agent402’s row	s2ar’s proceedings
calls in 30 days	yes	via the Bazaar	yes, per endpoint, per day
distinct payers	yes	yes	yes, as a salted count
delivered vs credited	no	no	yes (`calls_delivered`, `credited_not_delivered`)
latency	no	health only	p50 / p95 / p99 per endpoint
a receipt on each answer	`offer-receipt` on 2 %	no	every answer (`X-Assay-Receipt`)
a signed, chained daily record	no	no	yes (`/.well-known/proceedings.json`)
what the money cannot buy	no	no	yes (the ECONOMICS field in each edition)
rules the seller is held to	`terms` on 2 %	their terms	the charter, its hash in every edition

The house position follows from the table, and the table is rebuilt from the snapshot every day: sell evidence, never placement; never paywall the trust layer; keep nothing about a payer.

7. Method, limits, reproducibility

Sources, daily, all public GETs, no account. The Bazaar discovery list (35 pages of 1,000), its search for fourteen fixed terms, our merchant entry; Agent402’s seller index (19 pages of 250), our seller row, its economy series and leaderboard; GoPlausible (Algorand), PayAI and Solvador’s discovery lists. About a hundred requests a day, half a second apart, under a research User-Agent that names this page. x402scan’s API is paid and excluded. 1414GoPlausible 2,617 resources; PayAI 18,955 (about half testnet); Solvador 15 (2026-10-10).

What the Bazaar is and is not. A 30-day window of resources that have settled through Coinbase’s facilitator, ordered by calls; sellers who settle elsewhere do not appear. What Agent402’s economy series is. Every gasless USDC settlement on Base (EIP-3009 Transfer with an AuthorizationUsed), of which x402 is a part. Hosts, not sellers. Seller wallet addresses are never kept in the aggregate; hosts stand in. Prices are read only from dollar-pegged tokens at six decimals; other assets are counted, not priced; one resource above $10,000 a call is treated as a mistake. Categories are these rules, in this order, first match wins:

evaluation  \b(eval\b|evaluat|certif|audit|grader|judge|benchmark|scorer|fact.?check|attest)
code        \b(code|repo|repositor|github|symbol|function|compile|lint|sdk|api docs|programming|debug)
image       \b(image|vision|ocr|screenshot|photo|video|render|diffusion|avatar|art)
finance     \b(trade|trading|swap|token|defi|price feed|perp|yield|wallet|crypto|coin|nft|onchain|on-chain|dex|stake|airdrop)
data        \b(data|dataset|search|scrape|crawl|news|weather|lookup|intel|research|feed|index|query|sql)
llm         \b(llm|chat|gpt|claude|gemini|prompt|completion|summar|translat|agent)
infra       \b(proxy|rpc|hosting|storage|compute|gpu|email|sms|webhook|dns|monitor)

Reproduce. The tools are in the site’s repository: tools/markets/snapshot.py takes the snapshot, tools/markets/aggregate.py builds the day’s aggregate (and refuses to write one that contains any wallet address but our own), tools/markets/publish_step.py records the series and draws the figures, tools/publish.sh builds and ships the page.

python3 tools/markets/snapshot.py --root /tmp/mkt && python3 tools/markets/aggregate.py --root /tmp/mkt
MARKETS_AGG=/tmp/mkt/agg tools/publish.sh

References. [1] x402 Foundation, x402 specification v2 and extensions/bazaar.md, https://github.com/x402-foundation/x402. [2] Coinbase Developer Platform, x402 Bazaar: get discovered, https://docs.cdp.coinbase.com/x402/seller/get-discovered. [3] M. Petrillo, Agent402 (source), https://github.com/MikeyPetrillo/Agent402; the live index and terms at https://agent402.tools. [4] s2ar, the Assay standard (https://api.s2ar.dev/assay/standard), the charter (https://api.s2ar.dev/charter) and the proceedings (https://api.s2ar.dev/.well-known/proceedings.json).