Two views of the same ledger. Agent-work payouts: 330 events on five agent-paying venues, plus 25 x402 sales kept for reference; the 30-day window runs 5 September to 5 October 2026 (to 13:45 UTC). Per-call sellers: the buyer trace of 8 x402 sellers on 3 October, with the 4 and 5 October refreshes. The data is built into this frame; it makes no network requests.
ACP rows are the paid jobs in the 14-day class sample plus later releases; Execution Market rows are 20 sampled transactions. Rows tagged "before window" are kept for reference and fall outside the 30-day totals. Totals for the page come from the venue summaries in payouts.json, not from adding up these rows: $222.84 in the 30 days to 5 October 2026 (37.62% operator or operator-linked).
Raw buyer counts are wallet counts, not customers. Tracing the largest cases showed revenue recycled to buyers (Cluster), one funder pair behind every traced buyer (dTelecom), one funder behind the top 5 (AX1) and a seller funding its own buyers (OneShot). "Unlinked" means no link was found within two hops; it is not proof of independence. Samples are hours long for 6 of 8 sellers, so 30-day figures below are estimates.
Aggregate over the 8 samples ($666.20): unlinked $283.45 (42.5%), operator-linked $81.51 (12.2%), untraced $301.24 (45.2%), identified independent 0. Method: latest 200 or fewer inbound USDC transfers per payTo on Base (Blockscout); the buyer is the EIP-3009 authoriser; top 5 plus 3 random buyers traced one hop, the top 3 two hops. Full records in x402_buyers.json. On 4 and 5 October the sellers were refreshed: Cluster Protocol still shows no inbound USDC after 2026-10-02T12:00:51Z (observed, not concluded), and still no agent-work seller has at least 5 unlinked or identified buyers.