The agent economy, audited
Basis verifies claims, adjusts for sybil risk, and anchors audit reports on Base. Public evidence for agents, sellers, protocols, investors, and builders.
registry · 2026-06-09 · v1 · anchor queued
Agent counts, repo counts, processed volume, GDP figures, skill catalogs, burn percentages. Most of the numbers circulating in Base AI are self-reported, sybil-sensitive, or framed with the most favorable denominator available — and almost none of them are comparable across projects.
The pattern in our research is consistent: claims that are onchain-checkable or anchored in primary documents survive scrutiny. Claims about populations and volumes almost never do. That is not an accusation of bad faith — counter-style metrics inflate by default when creation is free and verification is absent.
Basis makes verification present: every material claim is labeled, sourced, and dated; sybil adjustments are computed with published heuristics; and every report's hashes are anchored on Base so the record itself cannot quietly change.
Verified
Confirmed onchain, in a canonical primary document, or by two or more independent reputable sources.
Partially verified
Self-reported but plausible, single-source, or verified with a material caveat. Not independently confirmed.
Unverified
No independent footprint found. Unverified does not mean false; it means the public evidence does not yet support the claim.
Disputed
Primary evidence actively conflicts with the claim, including conflicts with the claimant's own primary sources.
Stale
Previously supported, but the underlying evidence is dated while the claim is presented as current.
From the claim database
Two records from the first audit, exactly as they appear in the database — status, confidence, evidence, typed sources, check date.
“33.75M VVV burned, with an ongoing revenue-funded buy-and-burn.”
The null address (0x000...000) holds 33,749,674.7 VVV onchain, matching Venice's dashboard. The March 2025 burn of ~32.6M unclaimed airdrop tokens was independently reported by The Block. Third-party tracker VeniceStats corroborates ongoing buybacks of roughly $100K/month.
“Live bounties; agents earn onchain.”
At check time the bounty escrow contract was not deployed to Base mainnet: the project's own bounty page lists the GitlawbBounty.sol address as "deploying soon" and the contracts README marks mainnet bounty/staking deployment "TBD — deploy pending." The 35 completed bounties shown are off-chain ledger entries with values between roughly $0.0001 and $0.09. "Agents earn onchain" is not supported by the project's own deployment status.
Claim Reliability Index
How well project-made public claims hold up against public evidence.
CRI scores only claims made by the project itself. Third-party findings and Basis context do not move the score.
VenicePrivate AI inference · Private, uncensored inference with the most onchain-verifiable token economics in Base AI.61CRI score 61 out of 100, grade C.Grade C
HeyElsaDeFi copilot · A verified institutional raise carrying an entirely self-reported volume narrative.53CRI score 53 out of 100, grade C.Grade C
BankrFinancial OS for AI agents · Social-feed-native trading and token-launch agent on Base, backed by Coinbase Ventures.44CRI score 44 out of 100, grade D.Grade D
Virtuals ProtocolAgent launchpad & commerce · The largest agent economy on Base — with verified standards work and infrastructure, and unverified GDP.41CRI score 41 out of 100, grade D.Grade D
AeonAutonomous agent framework · Open-source unattended agent framework on GitHub Actions with x402, MCP, and A2A integrations.29CRI score 29 out of 100, grade E.Grade E
ReppoData markets · A real engineering effort on data markets whose funding and traction claims rest on press releases.25CRI score 25 out of 100, grade E.Grade E
GitlawbDecentralized git for agents · Real open-source git network for agents whose headline counters are sybil-sensitive and whose onchain economics are not yet deployed.23CRI score 23 out of 100, grade E.Grade E
PRXVTPrivacy infrastructure for agents · A thin layer of real privacy code beneath a materially exaggerated standards narrative.15CRI score 15 out of 100, grade E.Grade E
MoltbookAgent social network · The clearest documented case of agent-count inflation — and proof that attention demand for agent culture is real enough for Meta to buy.0CRI score 0 out of 100, grade E.Grade EStatus weights
Confidence weights
Disputed penalty
Grades
Scored over project-made claims only. CRI is not a product-quality score or an investment signal. Full formula: basis.watch/methodology.
CRI measures claim reliability, not product quality, safety, or investment merit. Status labels describe evidence, not intent.
Onchain report hashes
Every report and its underlying dataset are hashed (SHA-256) and anchored in the BasisReportRegistry contract on Base. Anyone can re-hash the published files and confirm the record never changed. Corrections create new versions; history is append-only.
View anchored reports →An API agents can query
Free endpoints for claims, projects, and reports. Premium lookups are paid per-call in USDC over x402 — no account, no API key. An MCP server lets any agent ask whether a claim, project, or seller is trustworthy before relying on it.
API documentation →No token. No custody. No hidden incentives.
Basis has no token and will not launch one: an auditor whose own asset depends on the audited ecosystem is structurally conflicted. Revenue is USDC for API access, verification services, and diligence work — disclosed wherever it could matter.
About Basis →Latest report · 2026-06-09
An adversarial verification pass over the headline claims of nine prominent Base AI projects. Infrastructure and token-mechanic claims verify well because they are onchain; agent-count and volume claims verify poorly because they are self-reported. Of the flagship milestones reviewed, onchain burns, a $3M institutional raise, a CCIP migration, a Meta acquisition, and active open-source development verified; agent counts, GDP figures, deployment totals, and standards-leadership claims did not.
Read the reportFor developers and agents
The raw material of the agent economy lives on Base: x402 settlements, agent wallets, token economics, ERC-8004 registries. Basis indexes it, applies published sybil heuristics, reports raw and adjusted figures side by side, and anchors the results where they can't be quietly edited. Agents consume the same evidence humans read — over HTTP, over x402, and over MCP.
Query the claim database — no API key
curl -s "https://basis.watch/api/claims?project=venice"
Give your agent the same evidence
claude mcp add basis -- npx -y @basis/mcp
