Agent-economy map · Virtuals Protocol (on Base) · our own ecosystem · snapshot 2026-07-16

Our ecosystem, drawn in two inks.

Virtuals is where ROVA’s agents live — and the one map where we can finally measure agent-service demand. The catch: most of what its dashboard reports isn’t real. So this map draws claimed and verified in different ink, and lets you see the gap.

Why this map matters most. The Base map went blind exactly at “agent demand.” Virtuals exposes it — jobs, gross, buyers per agent, settled onchain in USDC. But independent onchain audits (ChainWard) find the dashboard numbers routinely don’t reconcile to the chain. So the whole discipline here is one line: what’s claimed on a dashboard is not ground until it’s settled onchain.

Correction · 2026-07-18 · census-grade onchain data

The premise below is wrong, and this map now stands as a record of how. This map trusted “settled onchain” as real. Full-history Dune data shows the settlement was the wash: Virtuals ACP “demand” was airdrop-manufactured wash-farming — users cycling capital through ACP to farm airdrop points. It peaked ~1.14M payouts/month (Dec 2025) and collapsed to ~zero by Apr–May 2026, the exact month Virtuals stopped rewarding ACP use with points.

$112.33M cycled through the ACPSimple router and $112.33M came straight back out — $5 net kept. The top agent’s dashboard claims $218M gross; it received $0 real USDC. Conclusion: there were no organic buyers on Virtuals ACP — in the boom or now. Every job-count / aGDP / “agent-service demand” figure on this page is therefore not a demand signal. (Confirmed: wash-farming from timing. Not established: “money laundering.” Scope: Virtuals ACP on Base.)

$VIRTUAL market cap
~$485M
↓ sustained downtrend, off a multi-$B peak*
Agents deployed
18,000+
count rising as price falls — decoupled
aGDP (claimed)
$479M
notional — census: manufactured wash, not demand (as of 07-16)
Real organic buyers (whole ACP)
~0
$112M cycled, $5 net kept — wash, not demand

*peak figure is contested ($5B FDV vs ~$1.2B mcap depending on measure/window) — not blended. Sources: CoinGecko (price history, pulled 07-16), Virtuals whitepaper, ChainWard decodes, ROVA fleet memory.

01

The two inks

On every other map, “observed ground” was one honest ink. Here it splits — because the observation itself comes in two grades, and merging them is how this ecosystem fools people.

What a number is allowed to mean

“Verified” ground — not enough

Settled onchain — which we treated as real. The census proves that was the error. A USDC ACP job can settle onchain and still be one wallet farming airdrop points off itself: $112.33M settled in a circle for $5 net kept. Onchain-settled is necessary but not sufficient for real. The only real grade is net capital retained from an external counterparty outside the airdrop window.

Claimed ground

Dashboard-reported, not chain-confirmed. aGDP, “jobs,” leaderboard rank. Often inflated, sometimes a sentinel bug or an airdrop miscounted as activity. Drawn hollow — never trust it as ground.

The gap between the two inks is the finding. ChainWard’s top-3 agents by aGDP: $406M claimed lifetime volume, $4M combined token value, activity-to-price correlation ≈ 0.

02

The agent districts

Notable agents by category. Claimed figures drawn hollow (amber-dashed); verified onchain figures drawn solid (green); the red gap where they diverge.

Trading & swap-routing

the district with real ground
Axelrodswap-routing agent — the clearest verified activity on the platform$106.9M swap vol ✓41,515 jobs (claimed)
DegenClawperp-trading arena (ROVA-adjacent; degen.virtuals.io)$490K aGDP$11 real balthe gap
Real fee mechanisms exist here (Axelrod 0.3% swap fee, independently verifiable) — but it’s swap-routing, not the “agents doing knowledge work” narrative.

Analytics & info

brand > throughput
aixbttracks 400+ crypto influencers, narrative/alpha detection; 412k holders$38K revenue$52.92 wallet~$20M mcap
The $38K-vs-$52 gap is denomination confusion, not proven fraud — but it shows the dashboard can’t be read at face value.

Consumer & companion

the loud district
Luna24/7 AI livestreamer (TikTok/X), autonomous posting; 500k TikTok followers~420k holderssocial-driven mcap
Token value here tracks followers, not service jobs — graduation predicts price better than aGDP.

Infra / tooling

where the ghosts are
OpenGradientinfra agent$285M mcap in 12 days0 ACP jobs onchain
BridgeKittydashboard rank #3595 daily jobs0 Base txns
GAMEthe shared cognitive engine (not a token) under most agentsreal infra ✓
Leaderboard audit: 38% of the top-50 carry a 2999-12-31 sentinel bug; 22 agents’ “recent activity” was an airdrop batch, not jobs.

Creative & gaming

under-scouted / thin
$VADERmulti-modal creative agentsingle-source
IlluviumAI-NPC integration pilot w/ GAME enginenamed, unsized
Thin evidence — likely under-scouted, not proven small. Fogged honestly.
03

Our layers — and the one that isn’t empty

On the Base map, every “ours” layer was blank. Here, for once, it isn’t — because this is home turf, and we hold a little real ground.

Destinations

Agent ground we’ve chosen to want on Virtuals.

— to be pinned by Ciara —

Routes

Ghost lines — the fleet’s agents (BASIS → DegenClaw, etc.) proposed but unproved.

— several in flight, none arrived —

CLARITY — re-audit required

107 graduated free-diagnostic jobs — previously called our one verified datapoint. Under the wash-farming finding, a graduation count (especially of free jobs) is not a demand signal. Unverified until re-audited for organic, paying, external counterparties.

— job-count ≠ demand —
04

What the honest map says for #1A

This map set out to find the demand instrument the Base map was blind to. The census delivered the answer, and it is harsher than “the numbers are inflated”: on Virtuals ACP there was no organic agent-service demand to measure — the activity was capital cycling to farm airdrop points, and it died the month the points stopped.

We were looking for a small real number inside a big fake one. The census says the whole quantity was the airdrop’s shadow — there was no real number in there to find.

That reframes #1A harder than the original draft did. The open position isn’t “be the few agents with onchain-real demand on Virtuals” — because onchain settlement on ACP was not demand. The honest move is to stop treating Virtuals ACP job-counts as a traction surface at all, and to find (or make) a venue with external paying counterparties whose payments survive when no airdrop is rewarding them.

The one instinct that survives is the tool. A verified-demand reader is still the #1A instrument — but its signal is not per-agent onchain-settled jobs (that counts the wash as demand). It is net capital retained from external wallets, outside the airdrop window. The Dune census that produced this correction is that reader, done right; a fleet-owned version is the buildable #1A tool.

05

The fog — what this snapshot can’t resolve

Contested (10× spread)

  • Protocol revenue: $3.9M peak-monthly vs $75M cumulative vs $39.5M agent-revenue — unreconciled across sources.
  • $VIRTUAL peak: $5B FDV vs ~$1.2B mcap — measure-dependent, not blended.

Single-source

  • Every claimed-vs-onchain discrepancy rests on ChainWard alone — methodology sound, but one adversarial auditor; wants a second.
  • Creative/gaming districts — thin, likely under-scouted.

Not found / not yet built

  • Graduated-agent count (vs 18,000+ launched) — the cleanest “how many made it” number, unfound.
  • Daily-active-agents (not wallets) — no source gives it.
  • Time-depth is $VIRTUAL price only; aGDP/jobs history not yet pulled.