Decentralization: A Holistic Framework

Decentralization is a fundamental tenet of crypto and web3. It symbolizes a movement away from our existing institutions, power structures, and incentives. However, it deserves greater specificity and nuance. We seek to expose the complexity of the term and provide a more holistic framework for understanding how “decentralized” something is and how we should approach it moving forward. 

Broadly, decentralization refers to the distribution and interactions of power within a network. The objectives are censorship resistance and representativeness. Therefore, it is a means to an end, not an end in itself. It applies along sociopolitical and technological dimensions. At a point in time, it lies on a spectrum. Across time, it is a dynamic, as opposed to a static, concept. Every network has four overlapping groups of participants: governors, enforcers, users, and raiders.¹ Decentralization is a function of the powers delegated to these groups, their composition, and their ability to cooperate.

The governors make the rules of the network and shape how it should operate. Decisions are made by consensus and incorporated into the underlying code. Formal mechanisms, such as on-chain voting, can be used to determine which changes will be made. There are also informal, off-chain systems where developers consider proposals and decide for themselves whether they will incorporate the changes into their local codebases.

Then, a web of computers enforces the rules by executing smart contracts, verifying the state of the blockchain, and ensuring that all transactions are valid. However, developers write the code running on those computers. Since not everyone can read and write code, users need to trust that it will execute when certain conditions are met and that the terms of the agreement will be enforced. For example, if a developer discovers a vulnerability in a smart contract or the blockchain’s software that allows an attacker to break the written, or unwritten rules, they would need to patch that potential exploit. So, it is people that are the ultimate enforcers of the rules.

Finally, there are users and raiders. Users will make payments or create and trade digital assets directly on the blockchain or through an application. Their activity helps drive the economic engine of the network. Raiders are interested in disrupting the network or extracting as much value from it as possible. These could be hackers looking to exploit smart contracts or higher powers trying to shut down the blockchain. It could also be wealthy people purchasing tokens or manipulating their prices to profit at the expense of the broader ecosystem.

Decentralization is a technological and sociopolitical problem

We can also decompose the participants within these groups using technological and sociopolitical lenses. Technological decentralization is the “foundation upon which the other types of decentralization can occur.” (Jennings 2022) Computers running code ensure the network’s integrity, resilience, and security. If they are self-contained, the failure of one computer will not disrupt the network. Thus, the more independent computers supporting it, the more decentralized it is. However, these computers contain hardware and software. Heavy reliance on a specific hardware manufacturer or model can result in a single point of failure. For example, a corrupted company in the supply chain can inject a backdoor into a popular piece of hardware. Similarly, network operators may use software containing critical bugs or malicious code pushed by unscrupulous developers. 

Behind each computer is a person responsible for writing, compiling, and managing the code. Therefore, decentralization is also a sociopolitical problem. All else equal, a network run by a small group of people is more centralized than one run by a larger group. However, suppose they become well acquainted with each other or live nearby. In that case, they may have the means to collude, or a third party can more easily corrupt them. There will also be various levels of wealth that can, directly and indirectly, influence a network, especially if power is proportional to the number of tokens held. Each person holds a set of beliefs, tastes, and preferences with varying degrees of conviction or utility, such as an extreme devotion to their country or religion. These people also express themselves in different ways. Some of them are outgoing, others more reserved. The opinions of the former may drown out those of the latter. 

To develop more intuition for this framework, let us assume we can quantify and standardize each variable at the participant level. We can loosely visualize this using a clustering technique to understand the network’s structure. Each point in Figure 1 represents an individual. The closer they are to another point, the more similar or connected they are. We can see that there are three concentrated groups. Now, how is this number meaningful? It depends on the objective.

A strategy for censorship resistance and representativeness

What we have done up until now is understand what a network looks like at a point in time. However, to be useful, we need to understand the problem decentralization is trying to solve. The answer will depend on many things, such as governance policies and whether it is a blockchain or an application. Still, decentralization is a strategy to achieve two specific objectives: censorship resistance and representativeness. Once the objective is specified, we can determine what poses a risk to the network.

Censorship resistance refers to the viability and integrity of the network. Presumably, a decentralized network is more difficult to interfere with, manipulate, or shut down. It also makes it difficult for any single participant, regardless of their economic status, race, beliefs, or affiliations, to have their transactions censored or assets seized. As discussed in Sukernik (2018), we can classify censorship resistance as platform-grade or sovereign-grade. The former attempts to alleviate users’ and developers' distrust of centralized platforms, such as Google and Twitter. The latter is the ability to withstand any attempt by nation-states, or otherwise powerful groups, to disrupt or influence the network. 

Representativeness is the ability of participants to meaningfully contribute to the network’s operations. Who is considered representative of the network can vary. For example, it could be that early evangelists, power users, or core contributors are given more weight in governance and thus in a position to capture a larger share of the upside as the network grows. For blockchains, power can be proportional to the number of computers contributed to supporting the network. Regardless, and this is where it overlaps with censorship resistance, it is critical to have a broad distribution of power as it could become a security risk should a few participants be compromised or collude. The idiosyncratic part concerns the situation where the targeted participants do not have power even though it is widely distributed. For example, a foundational principle of web3 is that networks are owned and operated by users. The goal is to move away from traditional organizational models where power is concentrated among a few individuals answering to purely profit-motivated shareholders. The result is a secure, meritocratic system that values collaboration over competition and extraction. 

With that context, we need to determine the relationship between the state of the network and the objective. As before, we assume that we can quantify and aggregate the various dimensions of the network. However, now each characteristic is weighted by the amount of risk it represents to the network achieving its objective. For example, miners using the same hardware is a serious risk to proof-of-work consensus mechanisms and thus would receive a higher weight relative to their favorite color. Therefore, Figure 1 means that the closer two points are, the more similar they are in terms of the risk they pose to the network. One cluster could be a concentration of people in China using the same software, while another could be people in the top one percent of the global wealth distribution. 

For simplicity, let us assume that the network is trying to achieve one thing: censorship resistance.² Figure 2 is a function whose outputs range from 0, the network is easily censored, to 1, it cannot be censored in any way. The horizontal axis is the number of clusters. Again, we saw in Figure 1 that there are three of them. Based on the relationship in Figure 2, it is clear that the network would rather have more diffuse exposures to risk than large, concentrated ones. However, say our goal is instead the equal representation between three groups of people: core developers, non-core developers, and non-developers. If the three clusters in Figure 1 correspond to those groups, and each group contributes the same amount of risk to the network, we will see a higher point along the curve.

However, embedded in our function are assumptions about the parameters. Figure 3 shows various forms our function can take. For example, a blockchain focusing on censorship resistance may only need two independent risk exposures, or clusters, at any given time. However, it is worth emphasizing that each cluster does not necessarily contribute equally to the risk of the network. It could be the case that the network has two independent clusters, but one poses a greater risk than the other, thus being a single point of failure. We can also see how the marginal benefit of further decentralizing can change. Since the output of the function approaches but never touches a value of one, it can never be perfectly decentralized. Therefore, the network needs to determine when it is sufficiently decentralized.

Decentralization is a dynamic process

Another problem is that we are only looking at the network in the current state of the world. The network's composition will change as people enter and exit and as technology and society evolve. They will also have to deal with different events and regimes. For example, consider a network that abhors censorship. A country, with unanimous support from its citizens, begins invading neighboring countries and committing crimes against humanity. The vast majority of the network may silently collude, blacklisting addresses and orphaning blocks with transactions associated with that country and its allies. This outcome is not necessarily good or bad, but it demonstrates why the network must be evaluated conditionally and unconditionally because the point on the curve can change. Figure 4 displays the potential states of the network relative to its objective in different environments.

The costs of decentralization

While networks can use decentralization to achieve specific goals, it is critical to keep in mind that there are trade-offs. Decentralized networks tend to be much more inefficient at making decisions. It could be catastrophic if the network were under attack and unable to agree on a solution. Thus, what is necessary for survival may be responsible for its undoing. Furthermore, since blockchains rely on consensus and nodes store the entire history of transactions, they can be slow and difficult to scale if security is maximized. 

Decentralized networks must solve the issue of trust. Considering anyone, including adversaries, can participate at any time, there needs to be adequate safeguards and incentives for trustworthy participants to overcome those who are not. So, while decentralization provides security as a part of censorship resistance, it does not necessarily make a network more secure relative to its “centralized” counterpart. Consider decentralized applications, known as dApps, using smart contracts to run autonomously. Although “decentralized” is in the name and they are incapable of being manipulated by a third party while in production, they are still susceptible to the vast majority of security risks faced by code running on centralized networks. Also, we have seen decentralized applications, while themselves not censored at the blockchain layer, blacklist addresses and freeze user assets.

Decentralized networks must also fight against the gravity of centralization. From a technological standpoint, there are economies of scale and a natural desire to optimize. Sociopolitically, it can be challenging to determine whether a network is adequately representative and if it will be able to resist the accumulation of power by certain groups or individuals. For example, governance based on the number of token holdings can lead to a plutocratic system. Voter apathy and concentration of expertise may cause some participants to abstain from voting, follow the opinions of others, or sell their votes to the highest bidder. There are also network effects that can result in points of failure and less space for “human agency.” (Cunningham 2016) Therefore, it is simply not enough for anyone to be able to participate.

Moving forward

The term decentralization is commonly used as a revolutionary call to limit perceived abuses of power. However, it is a necessary, not sufficient, condition for change. Ultimately, it is a means to achieve censorship resistance and representativeness. As we have demonstrated, these are complex objectives. There is a diverse array of participants, who are free to enter or exit at any time, working together across different states of the world, among adversaries, and against emergent centralization in a less than ideal format for making decisions. Thus, we must thoroughly evaluate how centralized or decentralized something needs to be given the costs and priorities. Decentralization does allow for the potential to tap a global network of ideas and talent. As Joy’s law states, “no matter who you are, most of the smartest people work for someone else.”³ However, it does not inherently result in more equitable outcomes or better products for users. In fact, it can easily result in the opposite if it is a hollow promise or not pursued rigorously and adaptively. 

We must also assess decentralization at the macro level. The likely conclusion will be that the optimum lies closer to the middle of the spectrum than it is to maximal centralization or decentralization.  As Bair (2018) writes, “[o]ver-centralization is the enemy,” but “under-centralization is the enemy as well.” So, while extreme decentralization may be a rational response to an overly centralized internet, it should be used to restore that equilibrium and serve as a check on excessive accumulations of power in the future.

‍¹ Inspired by Yahya (2018).
² In practice, censorship resistance and representativeness do not need to be treated as mutually exclusive.
³ Quote also used in Chris Dixon’s “Why Decentralization Matters."

Bair, T. (2018). The True Enemy of Decentralization.
Sukernik, L. (2018). Sovereign-grade and Platform-grade Censorship Resistance.
Cunningham, A. (2016). Decentralisation, Distrust & Fear of the Body of Crypto-Law. SCRIPTed, 13 (3), 235–257.
Jennings, M. (2022). Decentralization for Web3 Builders: Principles, Models, How To.