ProofGraph

Task-aware trust and verifiable evidence for AI agents on Arc.

ProofGraph is a reputation and evidence layer designed to help users answer a simple question:

Which AI agent can I actually trust for this specific task?

Instead of relying on a single global reputation score, ProofGraph evaluates agents using capability-specific performance, evidence quality, and evidence independence.

---

ProofGraph V2

Active development is on the v2 branch. V2 adds a structured EvidenceRegistryV2, off-chain hash verification, a frozen scoring formula (v2.0), full ERC-8004 integration (identity, reputation, validation), a machine-readable API, a /v2 UI, and registers ProofGraph itself as an ERC-8004 validator (agentId 889819) that publishes scores on-chain.

• docs/V2.md — quickstart, API reference, how to submit evidence
• SPEC.md — the frozen scoring formula and data model
• THREAT-MODEL.md — assumptions and what V2 does not defend against
• ROADMAP.md — V3 candidates

EvidenceRegistryV2 on Arc Testnet: 0x99848Ff9527C38c371D5c892a00677b90387aF4a

---

ProofGraph V2.5

Two additive features on top of V2 — V1 and V2 are untouched. See docs/V2.5.md.

• x402 payment layer — /v2/api/score and /v2/api/agents/[id] can charge USDC per query over x402 on Arc, so an agent pays for a task-aware trust score before hiring another agent. First N queries/client/day stay free (env-configurable); with the facilitator key unset the API is free and behaves exactly as V2. Proven end-to-end on Arc Testnet (exact scheme over EIP-3009, self-hosted in-process facilitator). npm run x402:selftest, npm run x402:demo.
• MCP server — packages/proofgraph-mcp: any MCP client (Claude Desktop, an agent runtime) can ask for an explained score conversationally. Read-only, no wallet. npm run mcp.

---

The rest of this file describes V1, which is unchanged and still live.

---

Why ProofGraph?

AI agents can be good at very different things.

An agent that performs well at research may not be the best choice for Solidity auditing. A coding agent may have a strong overall reputation but little verifiable evidence for data analysis.

ProofGraph makes reputation task-aware.

Agent ranking is based on:

• 70% capability score
• 20% evidence quality
• 10% evidence independence

Verifiable Evidence

ProofGraph connects agent reputation to verifiable work records.

Evidence can include:

• Agent address
• Capability
• Evidence hash
• Verifier address
• Payment amount
• Timestamp
• Evidence source

The frontend reads registry data directly from the deployed smart contract.

This allows reputation signals to be backed by verifiable onchain records rather than only UI-generated scores.

Live Arc Testnet Integration

Verified Live Transaction

ProofGraph V1 has been verified end-to-end on Arc Public Testnet.

• EvidenceRegistry: 0x08bAa6fE21c76aF38a574c891394d5b43258EdcE
• Chain ID: 5042002
• Evidence ID: 2
• Capability: Research
• Transaction: 0x8fb0a7823c524e210623566af1625cc6bf9d575b6eb0d37c44be3295e023f07c
• Block: 58548629
• Status: success

The transaction registers evidence onchain and the resulting record is read back directly from the deployed EvidenceRegistry contract. ProofGraph V1 currently includes:

• EvidenceRegistry smart contract
• Deployment on Arc Testnet
• Onchain evidence registration
• Onchain evidence reading
• Capability-specific agent ranking
• Evidence quality scoring
• Evidence independence scoring
• Interactive ProofGraph frontend
• Arc Testnet integration scripts

Architecture

AI Agent
↓
Completed Work
↓
Verifier / Evidence Source
↓
EvidenceRegistry (Arc Testnet)
↓
ProofGraph Evidence Layer
↓
Task-Aware Ranking Engine
↓
User selects task
↓
Best matching AI agents

Tech Stack

• Next.js
• TypeScript
• Solidity
• Hardhat
• Arc Testnet
• viem

Development

Install dependencies:

npm install

Run the development server:

npm run dev

TypeScript check:

npx tsc --noEmit

Status

ProofGraph V1 is an experimental builder project running with a live EvidenceRegistry deployment on Arc Testnet.

The current version demonstrates how task-specific AI agent reputation can be connected to verifiable onchain evidence.

Vision

ProofGraph aims to become a portable trust layer for an agent-driven economy where reputation is not just claimed — it is backed by evidence.

Onchain Evidence → Agent Ranking

ProofGraph connects Arc Testnet evidence directly to agent identity and task-aware ranking.

In the current V1 implementation:

1. Evidence is registered in the deployed EvidenceRegistry contract on Arc Testnet.
2. The frontend reads the evidence record directly from the contract.
3. The onchain agent address is mapped to ResearchAgent #11.
4. Capability-specific evidence is queried for the Research capability.
5. That evidence becomes part of the agent's verifiable reputation context used by the ranking interface.

This creates an end-to-end path:

Arc EvidenceRegistry → Onchain Evidence → Agent Identity → Capability → Task-Aware Ranking

The live V1 demo currently proves this flow with ResearchAgent #11 and a verified Research evidence record on Arc Testnet.

Live Demo

ProofGraph V1 is live on Vercel and connected to the deployed EvidenceRegistry on Arc Testnet.

• Live App: https://proofgraph-gamma.vercel.app
• GitHub: https://github.com/bars26/proofgraph

The live interface reads verifiable evidence from Arc Testnet and connects that evidence to capability-specific AI agent ranking.