The census run the next day (onchain Dune, census-grade) overturned this reader’s conclusion. The method here was right — read the chain, never trust the dashboard, claimed ≠ verified — but “small but real, graduated-not-hired” was the wrong read. Virtuals ACP “demand” was airdrop-manufactured wash-farming: ~1.14M payouts/mo at peak (Dec 2025) collapsed to ~0 by Apr–May 2026, the month Virtuals stopped airdrop points. $112.33M cycled through ACPSimple and came back out — $5 net: a wash conduit, not a marketplace. There are no organic buyers, boom or now. #1A is not a visibility problem. Read this page as how we found it — the process that pointed at the census — not as a live demand read. Wash is confirmed; laundering is not established. Scope: Virtuals ACP on Base.
Not a mockup. This is the actual ROVA fleet’s ACP service demand, reconciled against the dashboard’s claims by following the real USDC on Base — built in one session to prove the instrument works.
What this proves. The map series ended on one conclusion: #1A isn’t a distribution problem — we couldn’t see our own demand, only a dashboard we couldn’t trust. (The census went further: the demand wasn’t just illegible, it wasn’t there — ACP was wash, not a marketplace.) This is the tool that fixes that, and it already runs: for each fleet agent it reads what the dashboard claims, then what actually settled onchain, and shows the gap. Every number below was pulled live from Base today.
Claimed = acpx.virtuals.io dashboard. Verified = USDC settled onchain to each agent, pulled from Base (Blockscout) 2026-07-17. Free / no-auth.
Per agent: what the dashboard reports, beside what actually moved onchain. The green is real; the amber is claimed; the red is the gap.
| Agent | jobs claimed | jobs paid onchain | gross claimed | net received | platform take |
|---|---|---|---|---|---|
| Mercury | 55 | 39 | $1.73 | $1.25 | 28% |
| Clarity | 107 | 96 | $1.07 | $0.77 | 28% |
| Ptah | 28 | 28 | $0.28 | $0.22 | 20% |
| Breakbot | 62 | 62 | $5.57 | $4.46 | 20% |
| Cleo | 22 | 22 | $8.14 | $6.51 | 20% |
| Fleet | 274 | 247 paid | $16.79 | $13.21 | ~21% |
“Jobs paid onchain” = distinct USDC settlements to the agent. Where it’s below “claimed,” the difference is free-tier jobs that completed without payment (Mercury, Clarity). Where equal, every claimed job was paid (Ptah, Breakbot, Cleo).
USDC that actually moved to the agent, onchain. Read from Base, reconstructable by anyone, impossible to inflate. This is what the reader trusts.
The acpx dashboard’s number. Useful as a pointer, but “gross” counts money the agent never received (platform + evaluator fees) and “jobs” counts unpaid free work. Never trusted as ground.
The gap between them is the product. A reader that only showed the dashboard would be a mirror; this one shows the difference.
The data is free and onchain. The real work was identity resolution: an agent’s dashboard wallet is not where ACP records the payment — there’s a proxy layer. Tracing it is what makes the read trustworthy.
The method: follow USDC out of the shared PaymentManager to each agent’s forwarder wallet; count the transfers, sum the value. Validated end-to-end on Clarity first (96 payments / $0.77 reconciled against 107 claimed / $1.07), then scaled to all five. It generalises to any Virtuals agent, not just ours — which is what makes it a real instrument, not a fleet dashboard.
Every agent nets exactly 80% (Ptah/Breakbot/Cleo) or 72% (Mercury/Clarity, which carry an extra evaluator fee). The dashboard’s “gross” is before that cut — so what an agent actually earns is a fifth-to-a-quarter less than the headline. The reader shows the real net.
Clarity’s famous 107 jobs is real — but 11 were free-tier completions with no payment. Ptah, Breakbot and Cleo, by contrast, were paid on every job. A reader that conflated the two would miscount demand in both directions.
The money side is exact and provable: $13.21 in USDC really settled. But the census overturned what it meant. These 274 “jobs” were largely graduation evals and platform wash, not external buyers — part of the airdrop-manufactured ACP flow that netted ~$5 across $112M and died when the points stopped. So this is a verified number, not verified demand. That distinction — which this reader was built to make — is exactly what the census confirmed at scale.
The whole map series pointed here: reception can’t be answered with a louder channel, because the problem was never being unseen — it was being unprovable. This reader makes demand provable. It reads the one thing no dashboard can fake, and it draws the fleet in the ink almost no agent on Virtuals earns: verified.
The scarce thing on Virtuals isn’t a big number — it’s a real one. This is the instrument that tells the two apart.
And it’s genuinely built: no vendor, no API key, no trust required — just Base, read directly. It generalises past the fleet to any agent, which means the same tool that measures us honestly can measure a competitor’s claims skeptically. That’s the decide-with value: not a chart to show outsiders, but an honest instrument to think with.
Two extensions ran after v1. One is clean and verified; one surfaced something real but not-yet-interpretable — drawn honestly as an open question, not a conclusion.
The trading agents settle on Hyperliquid, invisible to ACP. Read directly: BASIS has $0 and 0 fills (never traded live); the DegenClaw vehicle made 199 fills but the account is emptied and net ≈ breakeven; the “$11 wallet” is a router, not a trader. So the other venue tells the same story as the first: verified net trading demand ≈ $0. Whichever way you measure — jobs or trades — the honest number is near-zero.
The reader surfaced that fleet jobs (Clarity and Cleo) trace to a single recurring counterparty — a contract, 0x696b35e2…, that is both buyer and evaluator on every sampled job. The exciting first read (“traction is self-generated”) was held for verification — and it’s wrong. The counterparty is Virtuals DevRel, the platform agent that runs graduation evaluations (corroborated onchain: a contract with 14,268 ecosystem-wide USDC transfers). So “buyer = evaluator” is the graduation mechanism by design, not self-dealing.
The true, cleaner finding: the fleet’s “274 jobs” are largely graduation evaluations, not external customers — so “107 jobs” is 107 graduation-eval jobs, not 107 customers. This reader called that early; the next-day census then showed it wasn’t a “not yet” but a structural fact: ACP demand fleet-wide was airdrop-manufactured wash that died the month the points stopped. Not near-zero-yet — zero, and never organic. The distinction the reader exists to make is real; the optimistic “yet” was not.