The number on the ledger is real. As a measure of real economic demand, it is inflated by roughly 8–15×. Depending on the cutoff and the definition, public trackers disagree by an order of magnitude, 6 of the current top-10 buyers trace back to one self-funded loop, and a third-party tracker flags up to 81% of volume as gamed. Here is exactly how we checked — every claim has a clickable receipt.
x402 — Coinbase's open standard for paying for an API call over HTTP — is widely described with a single big number: ~$51M in cumulative volume. That figure is not fabricated. An explorer that sums on-chain USDC transfers will honestly arrive there.[1] The problem is what the number is taken to mean.
For the same 30-day window, different measurement sources reported volumes that differ by an order of magnitude — not because anyone miscounted, but because they measure different things: raw on-chain transfers versus volume with self-dealing and wash filtered out.
| Source | 30-day volume | What it measures |
|---|---|---|
| x402.org dashboard (via Bloomberg) | $24M | raw, unfiltered |
| Allium | $3M | raw on-chain |
| Artemis / a16z | $1.6M | wash removed |
$24M ÷ $1.6M = 15×. $24M ÷ $3M = 8×. An a16z researcher put the gap down to "how early the measurement infrastructure is."[2] Our reading: the raw headline is a real ledger total, but as a proxy for demand it should be read with a 10×-ish discount. The rest of this report is us doing that discount on-chain, by hand.
Walking the number down, layer by layer, each layer sourced:
The headline. Real on the ledger, but cumulative and unfiltered.
What is actually moving now. That annualizes to roughly $9–10M/year nominal — before removing any wash.
This strips economic demand — buyers paying for value. It deliberately does not deny that some removed flow is real infrastructure being load-tested: that's a genuine technical-adoption signal, just not paying demand. The band is wide because wash estimates genuinely are — we show the range, not a false-precision point.
Concentration alone proves nothing — a single heavy user can look like a whale. So we did not stop at "these addresses are big." We followed the money on-chain. On 2026-06-14 we pulled the top buyers by volume and traced their USDC on Base.[3]
Six buyer wallets, each credited by the tracker with ~$5,275 of volume and ~510k transactions, each touching exactly 3 sellers, each with a price-per-tx of $0.0103 — numerically too uniform to be six independent users. On-chain, they share one source and one sink:
Every farm wallet's incoming USDC comes from the same treasury; its outgoing USDC returns to the same cluster. The sink address shows 31.5M transfers and forwards what it collects back to the treasury. The money circles among related wallets — and each loop registers as "volume" on the leaderboard.
To prove the method isn't just "high volume = bad," we ran the same trace on the #4 buyer. It is the opposite of a loop: its USDC fans out to 942 different counterparties, never circles back to a single funder, and it holds real gas. That is a genuine multi-service probe or crawler — high volume, but real. The method separates the two.
Does a third party agree? Roughly. Artemis independently flags ~50% of transactions as gamed, which lines up with our own on-chain relationship checks (~54% by $). Its higher "~81% of volume" figure is a different lens (by volume, late-2025, Solana-heavy), so we treat it only as a directional second opinion, not a precise input.[4] Given the public sources disagree by 10×, our primary evidence is deliberately not anyone's filtered percentage — it's the specific, clickable clusters above that we can trace by hand.
public.buyers.all.list (sort by amount) → current top buyers0x8335…2913 token-transfers per address → in/out counterparties, shared funder_recon-topbuyer.mjs · _recon-verify.mjs · _recon-hub.mjstopbuyer-forensic-2026-06-14.jsonOne number here is easy to misread, so let's be precise. We pulled the current buyer universe (32,191 indexed buyers, ~92% of buyer-side dollars) and bucketed everyone by how many distinct sellers they actually paid.[5]
| Buyer type | Count | Share |
|---|---|---|
| Single seller only (sel = 1) | ~25,200 | 97% of buyers · 93% of tx |
| Multi-service (sel ≥ 3) | 392 | 1.5% of buyers · 38.5% of $ |
| Heavy multi-service (sel ≥ 5) | 226 | 0.9% of buyers |
"sel = 1" is not the same as "wash." That single-seller bucket mixes two very different things: wash farms hammering one endpoint, and perfectly real users who just need one specific API. Single-point usage is real value — it simply doesn't need a discovery or audit layer, because you already know who to call. We are not writing it off, and we are not claiming the protocol's only real demand is 392 addresses.
What 392 is: the buyers who genuinely shop across services (sel ≥ 3) — the ones who hit the "which of these is real?" problem an audit layer answers. That is Atlas's addressable market, not the protocol's total worth. It's small (stable vs our 06-06 read of 349), and concentration is extreme — a single address accounts for ~72% of all transactions.[6] An early, thin, top-heavy market — which is exactly why a wash filter is worth more than a bigger directory.
buyer-census2.mjs · agentic.market full buyer pull + seller-diversity bucketsbuyer-census-2026-06-14.jsonThe 2025 surge was real but largely a pay-to-mint meme phenomenon. It peaked and fell hard, and the "second wave" is a low, flat plateau — not a rising tide.[7]
| When | Daily tx | Note |
|---|---|---|
| Dec 2025 (peak) | ~731,000 | meme-mint era top |
| Feb 2026 | ~57,000 | −92% collapse |
| Jun 2026 (now) | ~$25–29k / day | low plateau, not climbing |
One apparent contradiction worth clearing up. Chainalysis says $1+ transfers are now ~95% of volume (up from ~49%) as sub-cent micro-experiments died off[7] — yet the wash cluster in ③ runs $0.0103 per tx. These don't conflict; they're two lenses. "95%" is by value — a few large transfers hold most of the dollars. The farm's penny transactions are by count — almost no dollars, enormous transaction count. Our own ticket-size split shows it directly: $1+ trades ≈ 86% of dollars but only ~5% of transactions; sub-$0.01 trades ≈ 40% of transactions and 0.5% of dollars.[6] The leaderboard ranks by cumulative amount, so 510k penny transactions still pile up to ~$5,275 and crack the top 10 — while contributing almost nothing to that "95% of volume." Both pictures are true at once.
The takeaway isn't "x402 is dead" — it's "the real market is small and quiet today, and the honest number matters more than the loud one."
0x3A0AA040…773E; every audit payment is a settled transaction anyone can open on Basescan. That is the whole point — an auditor you don't have to trust, because you can check.This report is the map of the market. The free onboarding report is the part you can act on: 6 services we paid for with real USDC — each with the exact call, the real response, and an on-chain receipt.
Your agent can buy the full paid-tested dataset over x402: GET /api/verified — 124 paid-tested services (111 verified + 13 flagged avoid), one JSON, $0.10. Verify, don't trust.