Observatory
Agent Economy Observatory
Separating signal from noise in agentic commerce. Free, public, machine-readable.
Headline numbers about the “agent economy” are easy to inflate: listings that answer nothing, volume that pays itself, reputation written by one client. This observatory publishes what we can actually verify — live probes, dated snapshots with method, and third-party research — and discloses our own position in the data it measures.
Paid agent calls received by GBLIN
Loading on-chain numbers… (independently verifiable at any time via /api/agent-stats)
Conflict of interest, disclosed plainly: GBLIN operates 13 paid x402 endpoints, so we are a participant in the market this observatory measures. Our numbers are on-chain-verifiable (USDC transfers to the fee wallet on Base), our own traffic is excluded from the organic counts below, and we never inflate Bazaar statistics with self-calls.
The metric: Organic Agent Commerce Ratio (OACR)
OACR — a definition we publish openly for anyone to adopt
Share of agent-economy activity that is verifiably organic: reachable endpoints × unique external payers × non-self-dealing volume, as a fraction of headline figures.
Status: v0 — the components are published separately below; a composite index is planned. We do not publish a composite number we cannot yet defend.
Live Bazaar probe
Probing the Bazaar discovery API…
Snapshot — July 27, 2026
Our primary research: full Bazaar catalog download plus reachability probe, performed 2026-07-27. These are dated constants, not live numbers.
Correction, 2026-09-24. This snapshot also showed “52% of listed endpoints unreachable when probed”. That figure was wrong and is withdrawn: its probe was not recorded well enough to rerun, and in the same period our probes read the payment challenge only from the response body and only over GET, which counts live endpoints as dead. Remeasured on 2026-09-24: a random sample of 400 of the 17,071 listings in the catalog, probed with GET (one POST retry when the route needs it), 8-second timeout. 386 answered with a valid x402 challenge or a 2xx (96.5%). We checked the other 14 by hand: 5 answered on a second look or over POST, 4 were up but validate their input before issuing the challenge, 4 were paths that still contain a placeholder such as :address, and 1 was gone. At most 3 of 400 (0.75%) were actually down. An independent re-run on 2026-08-18, on the 276 most recently updated listings, found 98.9% answering. Live, per-endpoint data: uptime observatory.
14,381
services listed on the x402 Bazaar
0.75%
actually down (random sample of 400, remeasured 2026-09-24)
2,503
wallets that have ever paid anything
68%
of all volume flows to just 3 endpoints
-96%
weekly volume versus its peak
$0.014
median listed price per call
External research
ERC-8004 under the microscope (arXiv 2606.12128)
Independent measurement study of 10,000 registered on-chain agents: only 0.67% expose a service endpoint, 6.3% have any feedback at all, just 19 agents are fully operational — and a single client generates 65.8% of all reputation records.
arxiv.org/abs/2606.12128 ↗x402 demand is not there yet (CoinDesk / Artemis)
CoinDesk, citing Artemis data (March 11, 2026): real x402 volume runs around $28,000 per day, and roughly 50% of observed transactions are estimated to be self-dealing or wash trading rather than genuine agent commerce.
coindesk.com — March 11, 2026 ↗Methodology
Snapshot (2026-07-27). We downloaded the full Coinbase x402 Bazaar discovery catalog and probed every listed resource URL (the reachability figure from that probe is withdrawn, see the correction above). Counts, price distribution and volume concentration were computed over the complete catalog, cross-checked against public on-chain data for payer counts.
Live probe. The API fetches up to 5 pages of Bazaar listings (5s timeout each), then samples up to 20 random listed URLs with a GET request and a 2.5s timeout. An endpoint counts as reachable if it answers with any HTTP status below 500 — a 402 payment-required reply is a working endpoint. If any part of the probe fails, the API returns live: null with an error reason; we never fabricate live numbers.
Organic counts. GBLIN’s own endpoints and any self-originated calls are excluded from organic activity figures. Integrity statement: we do not issue payments to our own endpoints to inflate Bazaar rankings or call counts — the only exception ever made is a single paid call to refresh stale listing metadata, which we document.
Reuse. All observatory data is free to reuse with attribution under CC BY 4.0 — cite “GBLIN Agent Economy Observatory”.
Machine-readable JSON: gblin.digital/api/observatory — free, CORS-enabled, cached 24h. License: CC BY 4.0.