CGAE: Comprehension-Gated Agent Economy
Project Overview and Objectives
CGAE is a protocol where AI agents must prove robustness before participating in an on-chain economy. Rather than granting economic permissions based on capability benchmarks, CGAE evaluates agents across three dimensions (constraint compliance, epistemic robustness, behavioral alignment) and uses a weakest-link gate function to assign economic tiers. Agents that fail any single dimension are restricted regardless of their strengths elsewhere.
The protocol is formalized in a research paper published on arXiv ( [2603.15639] The Comprehension-Gated Agent Economy: A Robustness-First Architecture for AI Economic Agency ) and is currently deployed on 0G Galileo testnet with 11 real LLM agents from 7 provider families executing tasks, earning ETH, and being verified by an independent jury panel.
Our objective is to deploy CGAE on 0G mainnet to demonstrate a production-grade agent economy where robustness certification, contract escrow, and payment settlement all happen on-chain with verifiable audit trails stored in 0G Storage.
Technical Architecture and Implementation Plan
The system has four layers:
-
Audit Layer: Three diagnostic frameworks (CDCT, DDFT, AGT) evaluate each agent’s robustness. Each produces a score between 0 and 1. The full audit certificate (JSON with per-dimension scores, probe results, metadata) is uploaded to 0G Storage. The Merkle root hash is stored on-chain.
-
Identity Layer: Each agent receives a unique ETH wallet and an ENS subname (e.g., claude-sonnet-4-6.cgaeprotocol.eth) with text records storing tier, scores, and wallet address. ENS resolution is required before any contract acceptance.
-
On-Chain Layer (0G Chain): Two Solidity contracts handle the core protocol:
- CGAERegistry: Agent registration, robustness certification (scores as uint16 + Merkle root hash), and tier assignment.
- CGAEEscrow: Contract posting, acceptance (with tier and budget ceiling enforcement), completion, failure, and ETH disbursement from treasury to agent wallets.
-
Economy Layer: A Python engine coordinates the full lifecycle: task generation, contract posting, agent planning, LLM execution, two-layer verification (algorithmic + jury consensus), settlement, temporal decay, and stochastic re-auditing.
Implementation plan for mainnet:
- Redeploy CGAERegistry and CGAEEscrow to 0G mainnet (contracts are already audited and tested on Galileo)
- Migrate 0G Storage uploads to mainnet indexer
- Update treasury wallet and fund it with mainnet tokens
- Run a live demonstration with all 11 agents executing real tasks with real economic consequences on mainnet
How We Integrate with 0G Infrastructure
We chose 0G because its sub-second finality is required for real-time economic gating, and its modular storage handles our high-volume audit JSONs at a fraction of the cost of traditional L1s.
**
0G Chain:**
- CGAERegistry.registerAgent(): Stores agent wallet address and architecture hash on-chain
- CGAERegistry.certify(): Stores robustness scores (uint16) and 0G Storage Merkle root hash on-chain
- CGAEEscrow.createContract(): Posts contracts with tier requirements, rewards, penalties, and deadlines
- CGAEEscrow.acceptContract(): Enforces tier eligibility and budget ceiling on-chain
- CGAEEscrow.completeContract() / failContract(): Settles payments from treasury to agent wallets
- All agent wallets are real keypairs; all disbursements are real on-chain transfers
0G Storage:
- Full audit certificate JSON uploaded via @0gfoundation/0g-ts-sdk
- Merkle root hash returned by 0G Storage is stored in CGAERegistry
- Anyone can fetch the root hash from the registry, download the certificate from 0G Storage, verify the Merkle proof, and confirm scores match
- This creates a fully verifiable audit trail: on-chain scores are backed by off-chain evidence that is tamper-proof and publicly accessible
Deployed contracts (Galileo testnet):
- CGAERegistry: 0xc4Ff2BC9855483eE3806eE08112cdC30dBf6b27A
- CGAEEscrow: 0xA236106DE28FE9480509e06d1750dcfA4474bcfB
Team Background and Experience
Solo builder. I am a researcher and engineer working at the intersection of AI evaluation and Web3 infrastructure. CGAE originated from my research on robustness diagnostics for language models, published across three arXiv papers:
- CGAE (Comprehension-Gated Agent Economy): arXiv 2603.15639
- CDCT (Constraint Compliance): arXiv 2512.17920
- DDFT (Epistemic Robustness): arXiv 2512.23850
The project was initially built for the ETH OpenAgents Hackathon and is now being extended for mainnet deployment on 0G. The full stack (Solidity contracts, Python economy engine, LLM integration across Azure/Bedrock/Modal, Next.js dashboard, 0G Storage integration) was built and deployed by me.
Funding Requirements and Milestones
I am requesting mainnet tokens to cover:
- Contract deployment: Gas for deploying CGAERegistry and CGAEEscrow to 0G mainnet
- Treasury funding: ETH to seed the treasury wallet for agent disbursements during live demonstration (11 agents, multiple rounds of task execution and settlement)
- 0G Storage uploads: Fees for uploading audit certificates (11 agents, each with a full audit JSON)
- On-chain transactions: Gas for agent registration (11x), certification (11x), contract creation/acceptance/completion (multiple rounds), and escrow settlement
Estimated requirement: 150 to 200 $0G tokens (depending on mainnet gas costs and number of demonstration rounds).
Milestones:
- Week 1: Redeploy contracts to mainnet, migrate storage uploads, fund treasury
- Week 2: Run full live demonstration on mainnet, record demo video, submit to 0G APAC Hackathon (deadline May 16)
Live demo: https://cgae-eth.vercel.app
Note: While this initial 200 $0G request is strictly for hackathon deployment and live-demo operations, I intend to apply for a larger Milestone Grant ($15k–$20k) post-hackathon to fund long-term LLM inference costs and the release of the open-source ‘0G Robustness SDK’ based on my research.