We read 59,000 DAO forum posts, so you don’t have to…*
* We used a language model 🤖 to find hot topics in DAO governance :)
Written by Christian Nielsen with input from Raphael Spannocchi
Overview
If you’ve ever sniffed a governance forum, you are probably aware of the myriad of discussions preceding all decision making in DAOs.
Well, we sifted through 59,000 governance discussion posts from across 15 of the DAOs we’ve engaged with to see if it could tell us anything about what DAO communities are consistently talking about and deciding on.
We wanted to find out: What are the most pertinent governance topics in DAO communities right now, and should you care?
The Need for Computational Discourse Analyses in DAOs
The volume of forum conversations can make it difficult to stay abreast of what is happening. As promising as autonomously ruled organizations are, today, most of the decision-making takes place off-chain through digital deliberation platforms, long before the first lines of resulting smart contract code are written.
The transparency and digitally-native proclivities of DAOs can turn the same overwhelming volume of discussions into a data goldmine. We have turned to the Forum data of 15 DAOs where Flipside Governance has contributed to identify some of the most salient governance topics and issues that are currently being discussed.
It is safe to say that the trustless transparency of DAOs makes for a lot of conversation data. although the numbers may not be useful in themselves, the fact that they are readily available for computational analysis means that broad discursive trends can be explored in a semi-automated fashion. Governance posts from 4,811 Discourse discussion threads from the 15 DAOs shown above were collected.
The conversations total 59,806 documents, each of which represents a member’s post or reply within a forum. Once preprocessed, the average word count of these documents is 789 characters or 160 words.
Through simple word counts and sum totals, we can already use this data to identify the absolute participation of our 15 DAOs. SeeFigure 2
The bars show the number of mean monthly posts by the different DAOs within the last three months. The communities are differently sized so the numbers do not represent the relative participation of DAOs. If this article intended to investigate relative participation, a representative chart could be produced by normalizing (or dividing) the monthly means through the size of the DAOs. The point here being that this data is uniquely valuable to an ecosystem whose decisions are primarily made discursively, not programmatically.
In the following sections we carve out one essential use of this data: to identify the most salient topics in DAO discourse right now. Crypto-newbies might find the topics informative in themselves, but the value of this article is more pronounced for crypto-native readers.
Speaking directly to the crypto-natives, we can say that if the topics found in the data are unsurprising and expected, it can only speak to the need for computational discourse analyses in each of the respective DAOs. It shows that discursive data can be computationally analyzed and synthesized so users can spend less time staying informed on Governance Issues and, therefore, spend their time on actual governance participation. This article only scratches the surface of the insights that can be drawn out from such research.
For those of you interested in the methods used for this initial analysis, you can find them at the end of the article. Other than that, let’s jump straight into our findings.
The most salient topics in DAO Governance right now
We identified two main areas of discourse in the data, each of which consists of two topics of discussion. The categories can be defined as:
- Issues and Solutions for Scalability
- Governance Participation
Within the first category we find discussion topics about (1a) Scalability Issues, (1b) Multi-Layer and Cross-Chain Interoperability. In the second category we found: (2a) Concerns Regarding Delegation and the Incentivization of Governance Participation and (2b) Larger Questions about Whether and How Delegates Should be Compensated.
1) Issues and Solutions in Scalability
2) Governance Participation
1a) Scalability Issues
1b) Scaling Solutions
2a) Concerns Regarding Delegation and the Incentivization of Governance Participation
2b) Larger Questions about Whether and How Delegates Should be Compensated
It goes without saying that a discussion of any of these four topics deserves an article of its own, if not a whole encyclopedia. However, for the purposes of this article, we will give an executive overview of the different categories and then finish up by briefly discussing the methods through which the categories were obtained, and their value.
Redacted topics + Qualitative Examples
Let’s briefly explore what these topics are about using some examples from the forum data.
1) Issues and Solutions in Scalability
The first broad category is one that has persisted in the DeFi and crypto space for years. Transaction fees and scaling were an obstacle even before CryptoKitties (temporarily) broke the network. This category touches on this long standing issue, but from the perspective of recent scaling solutions.
1a) Issues around scalability
The vast majority of posts in this topic deal with fees and gas prices associated with transactions that must still be performed on Layer 1. As such, the topic is mostly occupied with troubleshooting operations that may require an “an ETH L1 transaction, which means gas fees”. Alternatively, the topic also serves to remind users of this issue as a means of justifying the steps taken by protocols to overcome it. Uniswap, for example, discusses supporting “Uniswap’s multichain mission and [expanding] cross-chain experiences”.
1b) Scaling Solutions
Are a related topic to the previous example and the words associated with it have significant overlap. Protocols have taken different routes to scalability and this topic largely involves these solutions. This is exemplified by the following post, “Aave v3 was deployed following a multi-chain strategy: Polygon, Avalanche, Arbitrum, Optimism, Fantom, and Harmony”.
There are Layer 2 solutions on the one hand and on the other hand you have Multi-chain and Cross-chain solutions. Layer 2 solutions offer off-chain networks that map onto a blockchain (Layer 1). Multi and Cross-chain solutions have to do with the interoperability or interaction across blockchains. Whereas Cross-chain solutions connect different blockchains via bridges, multi chain solutions allow unconnected or compartmentalized blockchains to interact via wrapped assets that correspond to assets in another blockchain. Given the number of blockchains in the space and the relevance of scaling solutions this topic is and has been very relevant for some time. There have been recent concerns about the security of some of the approaches. Of course, DAO discourses reflect this balancing act between the desire for scalability and interoperability on the one hand and the risk trade-offs associated with these solutions.
2) Rapidly Developing Conversations Around the Issue of Governance Participation
The latter topics are more directly tied to governance or ‘meta-governance’ and pertain to the need for larger participation in DAOs. Decentralized organizations distribute more responsibility to individuals i, and although this structure is the foundation of DAOs egalitarian promise it requires involvement from members, like directly voicing their opinions or delegating it to third parties.
Within this topic we find questions regarding whether and how to pay delegates and self-delegation, soul-bound tokens and other solutions to questions of participation. Both these topics are additionally threaded by the potential dangers of centralization associated with such solutions.
2a) Concerns Regarding Delegation, and the Incentivization of Governance Participation
In the DAO ecosystem(s), delegation of votes has been a popular solution to issues of participation or effective governance. Of course, delegating power is, by definition, a centralizing practice yet it can also serve to mitigate problems in governance. For one, there are very convincing arguments as to why token-based voting leads to centralization especially when whales are factored into the equation.
Solutions to this issue have involved delegation, self-delegation as well as soul-bound tokens. To read more on the latter category we recommend this article . One user commented that “[there’s] an entire graveyard of projects and ecosystems that’ll show you that grants, and airdrops are in themselves not enough to sustain effort… What I’m saying is that sometimes you need to drag people to the party when you want them to show up. That’s what facilitating self-delegation is to [him]”.
Later, the same user acknowledges problems of collusion and centralization associated with self-delegation. On the other end of the spectrum there are comments which focus more on issues of centralization directly. These are often nested in a larger awareness that “token-weighted voting is far from perfect, but it is currently the fairest voting mechanism we have at a DAO-level…Issuing something like a soul-bound token for the purpose of governance invariably excludes people from participating.”
2b) Larger Questions about Whether, and How, Delegates Should be Compensated
A similarly relevant topic is that of delegate compensation, which is increasingly being considered as an incentive for participation and time commitment. Discussions around the topic are not limited to whether delegates should be compensated, which is largely seen as advantageous by now, but also to how this compensation should be determined.
For a more in-depth discussion on the topic check out this article.
We hope you enjoyed this executive summary of the four most salient topics being discussed right now! Earlier we noted that users in the DAO ecosystem may already have an expectation that these topics are being discussed. Yet, this very fact means that, through discursive analytics, these insights and more can be derived and analytical procedures can be refined. The discursive design of DAOs has a level of transparency and availability of data which remains largely untapped, and its analysis merits our attention. Let us explain how the humongous amount of forum data yielded these four topics.
Preliminary Computational Methods [C1]
The data was collected from Discourse, the most popular forum software for DAOs.
Discourse fora are split into categories whereby DAO members can address issues in accordance with the categories of discussion that are predetermined by the DAO. The segmentation was helpful for labeling data.
Using the forum architecture, I began my analysis of DAO discourses by identifying changes in category activity within each of 15 DAOs. By using the number of posts in a category as a proxy, I gauged which topics had seen the highest increase of activity in the past three months. As a result, the salience of the identified topics pertains solely to the past three months. This work, therefore, pinpoints the most salient governance topics in the past three months.
Using Flow as an example, the above bar-chart shows (1) the total activity for Flow discourse categories since the inception of Flow’s discourse forum, (2) the total activity for Flow discourse categories in the past trimester, and (3) the proportional change in activity for Flow discourse categories between the two time-periods (all-time Vs. past-trimester).
If a category has received 10 posts in the last trimester and it barely saw much activity in the past year, the bar for proportional change in activity will be large highlighting a difference in the number of recent posts relative to the number of posts it usually receives. In the above example we can see that the category of Delegate platforms experienced an explosion of activity in the past three months. By running the same analysis for each of the 15 DAOs, we identified the categories that had seen the most growth in the past three months.
Evidently, this first observation only goes so far and has limitations when attempting to understand the content of discussions within the categories; fortunately, techniques from the field of Natural Language Processing and the Digital Humanities can help shed light in this endeavor. Some of these methods involved:
- Netnographic exploration and classification of the individual categories
- Some of the explored categories are described in figure 5. Consequently, comments from each of these categories can be sampled, read qualitatively, and even classified through Machine Learning models.
- Early computational approaches to hermeneutical analysis of large text corpora through word counts.
- Through these approaches you can make. Figure 6 is a WordCloud, which shows cool visual representations of the most prevalent words in a corpus, for Aave’s development discussion category.
- More recent language modeling techniques that account for the relational character of language.
- One such technique uses Word-embeddings, which I found particularly useful. The model learns word associations to detect words with similar meanings based on Firth’s often cited contention that “you shall know a word by the company it keeps” (Firth, 1957). The word lists found in Figure 7 are based in the Word-Embedding model that I trained on the trimester corpus. Yet, to gain perspective on the significance of these lists I compared the Word-embedding trained on the past trimester to another Word-embedding that I trained on the whole corpus. To get a more detailed understanding of this approach, or to discuss my methods in more detail, please feel free to contact me, I did not want to flood this article with methods as I understand it is not a research paper but, in the spirit of open science, I have more detailed papers in the works and a website where I will publish data, analyses, and more!
Final Categories
In the final analysis, I was left with two broad categories each of which was split into two smaller topics. The table below depicts some of the most salient words, or the most commonly related to the most frequent words as modeled by the Word-Embedding I trained. To put it differently, and with less obnoxious jargon, the words in the last columns of the table were most related to the hottest categories, from which the four topics can be understood.
Conclusion: Why should the preliminary discussions deserve our attention?
Governance in all organizations is a balancing act. It refers to the system of rules and processes by which an organization is directed and controlled. Organizations that are sustainable over time must ensure that the interests of all stakeholders are considered. For example, in the world of corporate governance, there is a reliance on legal structures that resolve the “agency” problem: how to ensure that the interests of shareholders are well represented when these owners are not managing the day-to-day decisions of a corporation. In this analog environment, corporations rely on annual shareholder meetings where owners can make their opinions heard by voting on shareholder resolutions. Shareholders choose a board, making it accountable to oversee shareholder interests, the board chooses management which makes day to day decisions.
The corporate world has long recognized that discussions along the power structure must be regulated to prevent unfair influence of certain shareholders. As well, partly conditioned by analog systems, it had to prevent paralyzing the organization by having all shareholders opine on day-to-day decisions and micromanaging. That is, regulation served to centralize authority because it was seen as necessary for the survival of corporate organizations. Of course, other measures of centralization in corporate governance were far less necessary and all the more centralized. One need only look to institutional funds like Fidelity or Vanguard to understand that the shareholder´s delegation of voting was less concerned with centralization and agency than it was with entrusting your shares to the most economically savvy fund manager; centralization was fueled just as significantly by economic returns.
We find different, but similar, problems in DAO governance right now and the solutions we come to will be determined just as much as by those who voice their ideas as by those who vote. Herein lies the importance of following, understanding, and engaging with the discourses that traverse the DAO ecosystem. I think it was Mark Twain who, probably inspired by Biggie Smalls, said “history never repeats itself but it often rhymes”; the fact that the solutions that are developing in DAO governance converge with those in the corporate world, may be a sign of larger issues in the trajectory of DAO governance mechanisms.
Sociological theorists have long discussed, not whether, but the extent and speed with which centralizing mechanisms take hold whenever humans organize collectively. The project and aims of decentralization aren’t any different, except for the not-so-minor detail that this centralization can be the responsibility of transparent and communally agreed upon code (be it smart contracts or the systematized governance procedures that a DAO proposal must go through). That, to me, is the revolutionary fairness and effectiveness of the DAO project and it is precisely what is being debated right now.
Perhaps one of the things that this kind of data collection allows us to do is to understand when our off-chain discursive engagement will yield the most agentive changes and developments in how DAOs operate and, by extension, what DAOs are. DAOs are only beginning to agree on the structures and codes that facilitate cooperation and the way that these decisions are being determined is largely off-chain discussions and governance forum deliberation. By participating in the conversation, you help define the trajectory of DAO governance, the extent to which it borrows from past solutions to governance dilemmas (such as corporate gov), and the values upon which DAOs will be build (be it the egalitarian fairness of decentralized governance or new approaches to the efficient organization of economic organizations).
Essentially, the operative reach of a discussion surrounding vote delegation extends well beyond who you delegate your vote to, or what kind of allowances are entrusted to the people that we delegate our votes to. Staying informed and participating in these discussions tangibly defines the governance mechanisms that are adopted by DAOs over time and, consequently, the very definition of DAOs. Computational analyses of these discussions are one such way in which the process of discursive participation can be made less costly. In this article, I began a preliminary analysis of this data, but far more insights can be derived, so be on the lookout for more analyses like this on the Flipside Governance webpage.
If you found this research interesting and want to sponsor more good content, please consider delegating to Flipside Governance. Below are links for the protocols we are most active on.
If you are interested in joining the Flipside Governance team, we are always listening — DM @flipsidegov on Twitter.