ERC-8004 Reputation Registry — concentration measurement
Kcalbin LLC, 2026-08-22. Reproduce everything below yourself; trust none of it on our say-so.

WHAT THIS IS
A current, independent measurement of who actually writes feedback into the ERC-8004 Reputation
Registry, on Ethereum, BSC and Base. It is a follow-up to arXiv 2606.26028 ("Can Trustless Agents
Be Trusted?"), whose data ends 2026-05-13.

CONTRACTS (identical addresses on all three chains, verified via eth_getCode)
  IdentityRegistry    0x8004A169FB4a3325136EB29fA0ceB6D2e539a432
  ReputationRegistry  0x8004BAa17C55a88189AE136b182e5fdA19dE9b63
  feedback event topic0
      0x6a4a61743519c9d648a14e6493f47dbe3ff1aa29e7785c96c8326a205e58febc
      topics[1] = agentId (uint256)   topics[2] = reviewer address

HOW TO REPRODUCE (zero dependencies, Python 3 stdlib, public RPC, no API key)
  python3 collect.py base     https://mainnet.base.org        1300000 9000
  python3 collect.py ethereum https://eth.drpc.org             216000 5000
  python3 collect.py bsc      https://bsc-dataseed.binance.org 800000 4000
  python3 analyze.py base_feedback.json ethereum_feedback.json bsc_feedback.json

WHAT WE DID *NOT* REPRODUCE, AND WHY IT MATTERS
The paper's headline Sybil rates (73.5% / 59.2% / 90.6%) come from a shared-first-funder graph:
cluster reviewers by the address that first sent them native tokens. Reproducing that needs
per-address funding history for every reviewer. Public RPC cannot answer "who first funded this
address", and the block-explorer APIs that can now require a paid key. We therefore do NOT publish
a Sybil percentage. Everything reported here is measured directly from feedback events, needs no
heuristic, and can be recomputed from the commands above.