On July 28, 2024, the Compound community passed Proposal 289, empowering a group of five token holders to steal about $24 million from the CompoundDAO treasury. Of course the community didn’t intend to authorize a big heist. And the “Golden Boys,” the group of token holders who advanced Proposal 289, probably weren’t planning one. Proposal 289 sought to allocate 5% of the DAO’s treasury holdings for full-control by the Golden Boys in their yield-bearing vault. Many railed against the Golden Boys’ proposal as a prime example of a “governance attack,” manipulation of a DAO’s governance mechanisms with the goal of draining its treasury or consolidating control.
Whatever the Golden Boys’ intentions, how did this danger arise in Compound? Compound is one of the pioneers of decentralized token-voting, and has a Nakamoto coefficient of 17, making it more decentralized than many proof of stake networks. Yet the Golden Boys’ exploit exposes CompoundDAO’s chronic voter apathy. The Golden Boys were able to squeeze through Proposal 289 with less than 7% of the total COMP supply. That’s because many token holders simply chose not to vote — on average, only 51 out of Compound’s almost 5000 members vote on on-chain proposals. But this isn’t captured in existing metrics and notions of decentralization. That’s a big problem.
Popular metrics of decentralization today, such as the Nakamoto Coefficient or the Gini Coefficient, are token-centric. They focus on token distribution over addresses.
Consider two hypothetical DAOs, WhaleDAO and MinnowDAO. WhaleDAO has its tokens split evenly among 5 wallets. MinnowDAO, which has tokens evenly split among 100 wallets.
This isn’t captured in existing metrics and notions of decentralization. That’s a big problem
At first glance, we would assume that MinnowDAO is more decentralized than WhaleDAO. However, this conclusion doesn’t consider who controls these wallets in the first place. Suppose, for example, that in MinnowDAO, 70 of those wallets are owned by one whale, i.e., the same person or group. Then MinnowDAO would actually be more centralized than WhaleDAO, with a single person controlling 70% of the voting power.
This is exactly what happened in the case of the Golden Boys. The group’s “yes” vote of around 700k COMP was split amongst dozens of wallets, with many of the individual wallets holding less than 25k COMP. Existing, token-centric metrics of decentralization, such as the Nakamoto coefficient, would lead us to falsely conclude that this vote was decentralized, since the “yes” vote was split across multiple addresses.
Token-centric metrics miss other forms of centralization too. What if 80% of voters in a DAO have been bribed to vote for a proposal? Bribery doesn’t affect token holdings, so a token-centric metric won’t capture it. The same is true for groupthink, collusion, and other ways in which control of a DAO’s governance may fall to a single or small number of groups.
To measure decentralization in a way that matches meaningful notions of the robustness of a DAO’s governance, there’s a crying need for a new metric.
To better understand the true degree of decentralization in a DAO, we propose a new family of metrics called Voting Bloc Entropy or VBE (pronounced “vibe”) for short. As the name suggests, VBE does not just measure token holdings in individual wallet addresses. It takes into account clusters of token holders that are exhibiting similar voting patterns. We call such a cluster a “voting bloc,” by analogy to voting blocs in political systems. Voting blocs may be loose coalitions or they may be self-identifying groups of individuals like the Golden Boys. VBE measures whether there are large, dominant voting blocs — what we regard as the key form of centralization in DAO governance.
In the case of the Golden Boys, a VBE metric would treat all the attackers’ votes as a single voting bloc, regardless of how many wallets they split their tokens into. It would also treat all of the non-participating voters in CompoundDAO as a bloc, since they too exhibit aligned behavior. Those voters, the majority by voting weight, would constitute a single, massive bloc that we call an “inactivity whale.” That huge bloc which the Golden Boys exploited — would imply low VBE and thus poor decentralization in Compound.
One of the most important benefits of a good decentralization metric is that it leads to useful, actionable guidance about how to reinforce decentralization. Our recent research enumerates a number of key lessons that VBE offers on DAO decentralization. For example, you might expect delegation to increase centralization by concentrating voting power. Surprisingly, our results show that when a DAO has a big inactivity whale, delegation can shrink the whale and lead to higher VBE, i.e., higher decentralization.
To illuminate the state of decentralization across the DAO landscape, we’ve created a new resource for the community: The IC3 VBE dashboard, which offers ways to compare VBE across DAOs (although some caution is warranted in cross-DAO comparisons) and enables DAOs to monitor fluctuations in VBE over time. As a proxy for the “health” of a DAO, VBE can help DAO operators guide governance practices.
We also believe that our research can inform the delegation process in DAOs. For example, by clustering delegates into voting blocs that resemble political parties, VBE offers token holders a way to easily identify delegates (possibly create AI agents) that can represent their interests, and in turn reducing voter apathy in DAOs.
With tools such as VBE, DAOs can better avoid governance attacks exploiting DAO voter apathy like the recent Compound case. Beyond this goal, though, by measuring and seeking to optimize their VBE, DAOs can achieve better governance that leverages a diversity of viewpoints in voting communities and fosters healthy debate. While our research is still a work in progress, we have involved major DAOs and stakeholders in implementing, testing, and iterating on an open-source library to power the dashboard and other applications of VBE.
As in so many other decentralized systems, the “D” in DAO is perhaps the hardest letter to get right. But maybe good vibes can help.
Note: The views expressed in this column are those of the author and do not necessarily reflect those of CoinDesk, Inc. or its owners and affiliates.