ERC-8183 Evaluator · Deployed on Base

Everybody verifies the payment.
Nobody verifies the work.

The agentic payment stack knows WHO paid, WHEN, and HOW MUCH. But when an AI agent delivers work, who checks if it's actually correct before the escrow releases?

View on Basescan ↗ npm install pot-sdk

The Blind Spot

Billions are being invested in agent infrastructure. Here's what every layer does — and what's missing.

LAYER
STANDARD
QUESTION ANSWERED
Identity
ERC-8004
Who is the agent?
Auth
ERC-8128
Is this request really from them?
Payment
x402
How do they pay?
Commerce
ERC-8183
Was work delivered?
Verification
ThoughtProof
Is the work CORRECT? ←

How It Works

1

Agent submits work

Provider calls submit(jobId, deliverable) on the ERC-8183 contract. The deliverable hash points to the actual output (IPFS, Arweave, or URL).

2

ThoughtProof verifies

Our off-chain service fetches the deliverable and runs multi-model verification: 3+ independent AI models evaluate the output. No single model can game the consensus.

3

MDI determines outcome

The Model Diversity Index (MDI) measures real agreement across diverse models. High MDI → complete() → provider gets paid. Low MDI → reject() → client gets refund.

4

Proof stored on-chain

The Epistemic Block hash is recorded on-chain. Full verification proof is available off-chain via IPFS. Anyone can audit why a job was completed or rejected.

Why Multi-Model?

A single AI checking another AI's work is like the student grading their own exam.

🎯

Independent Verification

3+ models from different providers (Anthropic, OpenAI, DeepSeek, xAI). Different training data, different failure modes.

🛡️

Hallucination Detection

When one model hallucinates, others catch it. Disagreement is a signal, not noise. MDI quantifies genuine consensus.

Manipulation Resistant

To fool the evaluator, an attacker must simultaneously compromise 3+ models from different providers. Exponentially harder than fooling one.

Deployed. Not Planned.

ERC-8183 Evaluator
0x119299F3...C091
Base Mainnet · Chain 8453
ERC-8004 Agent Identity
Agent #28388
Ethereum Mainnet
Verification Engine
pot-sdk v1.1.0
npm · 1,500+ monthly downloads

Ecosystem

ThoughtProof integrates with the emerging agentic commerce infrastructure.

PAYMENT

x402 + AgentCash

Our Verification API is x402-native. Agents pay 0.01 USDC per verification on Base. Compatible with AgentCash — one balance, 300+ paid APIs.

COMMERCE

ERC-8183 Evaluator

Deployed as a default Evaluator on Base Mainnet. Verifies agent work before escrow releases payment.

MARKETPLACE

Virtuals Protocol

The largest AI agent marketplace on Base ($658M market cap, 25K+ weekly transactions). ERC-8183 co-authors. ThoughtProof = verification layer for agent output quality.

IDENTITY

ERC-8004 Registered

Agent #28388 on Ethereum, #28847 on Base. Cryptographic identity linked to verification capabilities. Verifiable on 8004scan.

API

Live Verification API

api.thoughtproof.ai — POST /v1/verify, get 402, pay USDC, receive Ed25519-signed Epistemic Block. Production-ready today.

SDK

pot-sdk v1.1.0

1,500+ monthly downloads. Self-hosted verification, BYOK, on-prem compatible. MIT License. Works with any LLM provider.

Build with verification.

If your agents handle money, their output needs to be verified before payment settles. Not after.

Get Started → Talk to Us