AgentRankr

AgentRank is an explorer for the emerging ERC-8004 agent economy on Ethereum mainnet. It ranks autonomous agents by an on-chain reputation score computed entirely in BigQuery SQL over raw chain logs — decoding the Identity Registry (ERC-721 mints) and Reputation Registry feedback events directly by topic hash, then scoring each agent by feedback volume, score distribution, and recency. The result is a searchable, filterable leaderboard of 34,000+ registered agents, each with a detail page showing reputation, owner, and metadata. Agents that accept x402 micropayments are flagged from their tokenURI registration files. Owner addresses reverse-resolve to ENS names with ENSIP-25/26 agent text records surfaced where set, and a live ENS resolver resolves any name or address against mainnet in real time. The reputation SQL is the product — the UI never touches raw logs.

The data layer is pure BigQuery SQL against bigquery-public-data.crypto_ethereum.logs. We derived the ERC-8004 event topic hashes from the official registry ABIs and decode Identity + Reputation registry events in SQL, computing a recency-weighted reputation score (feedback volume, score distribution, time decay) in one CREATE OR REPLACE TABLE that materializes agentrank.agent_summary — bounded to a 400GB billing cap and kept fresh by a native BigQuery scheduled query. The web app is Next.js 14 (App Router) reading the materialized snapshot, so judges never wait on a cold 30s scan. A tokenURI/IPFS metadata fetcher with server-side cache detects x402 support. ENS is real on-chain resolution via viem through the ENS Universal Resolver (forward + reverse + text records), never hard-coded. The signature visual is a p5.js force-directed "reputation flow" where node color encodes score and size encodes feedback volume. Deployed on Google Cloud Run (Next.js standalone container) with a Vercel fallback.