Hi
A note reports the method and on-chain results of deploying an ERC-8240 reference standard under multisig governance across two ceremonies on Base mainnet, structured as an instance of its own read-back protocol. It lives, versioned, in the agents-public repository (agents-public/notes/birth-of-a-governed-standard.md at 850e1ef59b256d6ddcb7ca69d023e02733382807 · nicobernad/agents-public · GitHub) and implements ERC-8240, submitted as ethereum/ERCs PR #1705
(Add ERC: Trust Infrastructure for Agents and Assets by nicobernad · Pull Request #1705 · ethereum/ERCs · GitHub). The stance is the corpus rule, the reader is not asked to trust any claim; the reader is asked to re-read it. Section 7 is a reproducibility table (statement, location, method, expected value), replayable with public endpoints alone , most rows with a single Base RPC / and the dormant state / attestation threshold 1, zero sealed transitions since genesis is published to be checked, not asserted. Corrections and independent re-reads welcome.
1 Like
Following the exchange between @Nicopat and @RNWY on BehaviorMonitor and DependencyGraph with interest —- the “ sibil alerts on- chain” / “hasSybilLink verifiable by any protocol” framing is exactly the layer I’ve been working on, from a different angle.
An empirical study published this year (Xiong, Li,Wei, Wang, Knottenbelt, and Wang —- “Can Trustless agents be trusted? An Empirical Study of the ERC-8004 Decentralised AI Agent Ecosystem,” arXiv:2606.26028) crawled on-chain feedback data across Ethereum, BSC, and Base and found 73.5% / 59.2% / 90.6% of reviewers respectively show coordinated Sybil behaviour, with the reliable rating pool collapsing for most agents on two of the three chains once the Sybil-flagged feedback is removed. That’s the concrete version of the problem BehaviorMonitor and DependencyGraph are aimed at
I built a small reference implementation exploring one specific structural approach to detecting this: combine stake-weighted bonding with two independent graph signals —- local conductance( fast, catches obviously isolated fake clusters) and a SybilRank-style trust propagation pass seeded from staked agents( slower, but catches “bridge attacks” where a Sybil ring adds a couple of edges into the honest network specifically to fool a one-hop conductance check). The combined suspicion score takes the worse of the two signals, so an attacker has to defeat both simultaneously. Each component (stake tier, conductance suspicion, propagation suspicion) is exposed separately rather than collapsed into one number, similar in spirit to RNWY’s “shows its math and leaves the decision to the consumer” posture.
Code, adversarial tests ( including a bridge-attack construction against the combined defence), and whitepaper here: GitHub - kenjimoto999/veylux · GitHub Open, MIT-licensed, no ticket.This is narrower in scope than the Assets half of ERC-1705 —- purely the agent-attenuation side —- but if the DependencyGraph/hasSybilLink works looking for structural detection approaches to compare against, genuinely curious how this holds up against the coordinated-cluster pattern you’re already seeing in production.