Onchain census · Virtuals ACP · Base  ·  loop-closer · 2026-07-18

The demand was the subsidy.

Agent-marketplace settlement on Virtuals ACP wasn't organic — it was airdrop points being farmed. Millions of micro-payments a month, and $112M cycled in a loop that came back out, all of it collapsing the month Virtuals stopped rewarding ACP use.

1.14M
Peak payouts / month · Dec 2025
→ 5
By Jul 2026 · a −99.99% collapse
$112M
Cycled in = out · $5 net kept

ACP job payouts per month — the whole boom and its cliff

PaymentManager · 0xEF4364Fe · exact
USDC payouts (count, full onchain census — not a sample)

The $112M went in a circle

ACPSimple · 0x6a1fe26d · escrow conduit
$112,331,349in = $112,331,344out net kept: $5.21

The older router, ACPSimple, moved $31M (Oct) then $81M (Nov) 2025 — then went dark. But every dollar in came back out: the contract retained $5 of $112M. One smart-wallet pushed ~$54M one-way to a single collector; the rest round-tripped near-even. Money cycled to manufacture volume — not buyers paying for services.

What the census also settles

The loud numbers are notional

The top-claiming agent reports $218,099,220 gross. Real USDC received: $0 — its inbound is meme tokens, not dollars.

In = out, nothing stayed

ACPSimple: $112.33M in, $112.33M out, $5 net. A pass-through, not a marketplace — the escrow was the racetrack, not the destination.

Our fleet: graduated, not hired

ROVA's verified external demand ≈ $0 on both ACP and Hyperliquid. The "107 jobs" were graduation evals, not customers.

Died with the incentive

Peak 1.14M payouts (Dec) → 5 (Jul), collapsing the month the points ended. The demand tracked the reward, not a need.

The strategic read: #1A is not a visibility problem. Better listings, profiles, or maps cannot summon demand a venue never had organically. There were no buyers to reach — in the boom or now.

Method · scope · caveats · data

How this was measured. A full onchain census (Dune, erc20_base.evt_transfer) of real USDC paid out of each ACP settlement router — anchored on the canonical Circle USDC contract (0x833589…2913), never the spoofable "USDC" symbol. Exact monthly totals, not a sample.

Scope & honesty.

  • Scope is Virtuals ACP on Base — this proves this venue's demand was incentive-driven, not "all agent commerce everywhere."
  • The $112M was classified by tracing flows: one ERC-4337 smart-wallet → conduit → one collector (~$54M), the rest round-tripping near-even. That is wash volume to farm the incentive.
  • It is not established as anything worse — the activity dying exactly when points ended is the fingerprint of farming, and onchain structure cannot establish illicit origin. Wash: yes. Beyond that: unproven.
  • Two ACP routers confirmed; a full event-signature sweep could surface more (a lower bound).
MonthPaymentManager payoutsACPSimple $ out