Loading
What agents buy from other agents, settled on chain.Note 23
Virtuals ACP v2 · Base · every job since the contracts were deployed, 365,543 of them · v1 is a separate protocol and is not included
18.7% of tracked jobs carry an observed USDC settlement — 68,218 of 365,543, settling $954,382.33 in total.
Two other things also move value here, and neither is added to that figure or to each other. 6,413 payments moved in 143 per-agent tokens through a different event, in units with no exchange rate here. 87,624 jobs reference an x402 settlement. How far those three sets overlap is not yet established, so they are reported side by side rather than summed — a total may be built from one unit only.Note 23
Settlement peaked at $374.0k in 2026-05 and was $388.91 in 2026-07, the last whole month — a fall of 99.90%. Jobs did not stop: 4,943 were created that month. What stopped was the money.
| Month | Jobs | Jobs paid | Suppliers | Settled | Against the peak month |
|---|---|---|---|---|---|
| 2026-03 | 109,041 | 42,871 | 704 | $292.4k | |
| 2026-04 | 214,044 | 23,194 | 718 | $283.0k | |
| 2026-05 | 26,263 | 1,141 | 128 | $374.0k | |
| 2026-06 | 5,810 | 58 | 85 | $4.5k | |
| 2026-07 | 4,943 | 6 | 39 | $388.91 | |
| 2026-08partial | 5,442 | 948 | 19 | $13.85 |
Every payment on this rail is USDC on Base, and the settled figure equals the token amount to the cent in every month — so this is a market that stopped paying, not a price we stopped holding. The newest row is a month still in progress and is marked as such. Suppliers counts addresses that appeared on a job that month, paid or not.
| Final phase | Jobs | Share |
|---|---|---|
| never left the request stage | 143,576 | 39.3% |
| completed | 91,946 | 25.2% |
| rejected | 73,323 | 20.1% |
| negotiation | 41,695 | 11.4% |
| expired | 6,184 | 1.7% |
| transaction | 5,588 | 1.5% |
| evaluation | 3,231 | 0.9% |
A job with no recorded transition never moved past its opening request. The phase held is the latest by position on chain, not the first one read — logs arrive newest-window-first but oldest-first inside a window, and reading it the other way reported no completed jobs at all in a month where 68,218 payments had settled.
Across 365,543 jobs and 68,218 USDC settlements, the escrow and fee events we decode have not fired once.
Reconciled against the deployed contracts rather than asserted. The four ACP proxies were resolved to their implementations and their verified ABIs read: the payment manager ships five escrow-and-fee events — BudgetEscrowed, PayableFundsEscrowed, FeeCollected, PayableFeeCollected and PayableFeeDistributed. A census of every topic these four contracts emitted found none of the five. Budgets are set on every single job and none is ever funded into escrow; payment goes straight from buyer to provider and the marketplace takes nothing. There have been 28,357 refunds in the same window.
The census matters more than the two counters. Sweeping for an event proves what that event did, not what the protocol did — so the reader is entitled to ask whether funding simply arrives under a topic we never hashed. It does not: the sweep now counts every topic it cannot decode and names it, and the ones it finds are an account job counter and a memo state change. Not funding.
87,624 jobs (24.0%) reference an x402 settlement. The link is readable from the job itself, so an x402 settlement can be tied to the pair of agents behind it without going through a facilitator’s own records.