Skip to content
Crypto.world

Research

Characterising Global Platforms: Centralised, Decentralised, Federated, and Grassroots

Informal labels like "centralised," "decentralised," "federated," and "peer-to-peer" are widely used for global platforms, but the paper argues they are imprecise: boundaries between classes are…

Characterising Global Platforms: Centralised, Decentralised, Federated, and Grassroots
arXiv
2511.03286
Published
2025-11-05
Authors
Ehud Shapiro

AI summary

Overview

Research area: Distributed computing and formal methods — specifically multiagent transition systems and protocol specification — applied to the classification of global digital platforms (social networks, cryptocurrencies, federated social media, grassroots systems).

Technical level: Advanced. The paper is built on formal definitions, transition-system semantics, an appeal to Zorn's lemma for well-definedness of essential sets, and propositions proved in an appendix. The prose framing is accessible, but the core machinery is not beginner material.

Scope: The paper proposes atomic-transactions-based multiagent transition systems as a single formal framework in which every global platform — existing or imagined — can be specified and then classified as centralised, decentralised, federated, or grassroots according to the cardinality of its minimal set of "essential agents."

What This Paper Is About

Informal labels like "centralised," "decentralised," "federated," and "peer-to-peer" are widely used for global platforms, but the paper argues they are imprecise: boundaries between classes are vague, the classification is not exhaustive, and it is unclear whether the classes are mutually exclusive. The goal is to replace this with a mathematical characterisation. The author introduces the notion of essential agents — minimal sets of agents whose removal makes multi-party communication impossible — and shows that the cardinality of such sets sorts all global platforms into exactly four mutually exclusive classes, illustrating each with formal transaction-based specifications of a social network.

Key Contributions

  1. A formal framework for specifying platforms. The paper proposes atomic-transactions-based multiagent transition systems and protocols as the vehicle for studying global platforms, and adds several new notions to that framework: transaction equivalence (grouping transactions that represent "the same action" in different configurations), liveness via a designated set Λ of live equivalence classes, and the notion of correct (live) runs.

  2. Essential agents as a classification lens. Essential agents are defined as minimal sets E such that for every population P disjoint from E, every run of the protocol has only unary transitions. The paper proves (Observation 11) that the minimum cardinality κ of such a set partitions all protocols admitting an essential set with ∅ ⊊ E ⊊ Π into four classes: centralised (κ = 1), decentralised (1 < κ < ∞), federated (κ = ∞ with some essential set having |Π ∖ E| ≥ 2), and grassroots (every essential set with ∅ ⊊ E ⊊ Π has |Π ∖ E| = 1). Canonical examples given are Facebook, Bitcoin, Mastodon, and Scuttlebutt respectively.

  3. Four specifications of the same social network. A Facebook/X-like social network with feeds and followers is specified four ways — centralised, decentralised (Bitcoin-like), federated, and grassroots — using multiagent atomic transactions, with propositions stating that all four satisfy the same Social Network Correctness properties while having different essential sets as expected.

  4. Reconciliation of two prior grassroots definitions. Grassroots platforms have two previous definitions in the literature. The paper proves that (1) the two definitions are incomparable, (2) each implies that all agents but one are essential, and (3) both implications are strict, by presenting two platforms in which all agents but one are essential but which are not grassroots under either definition.

Main Findings

  • The four-class partition is exhaustive and mutually exclusive. Observation 11 proves that exactly one class applies to any transactions-based multiagent protocol admitting an essential set E with ∅ ⊊ E ⊊ Π: if κ < ∞ then the grassroots condition fails and exactly one of centralised or decentralised holds; if κ = ∞ then neither holds and exactly one of federated or grassroots holds, since those two conditions are complementary.

  • Centralised platforms have exactly one essential agent. For the centralised social network specification, Proposition 15 states E = {p} and |E| = 1, where p is the named server. The server is required for any transaction to occur: Register and Sync have p as a participant, and Post and Follow require prior registration.

  • Decentralised platforms have a finite, greater-than-one number of essential agents. For the Bitcoin-like specification, Proposition 19 states E = B and 1 < |E| < ∞, where B is the set of bootstrap nodes hardcoded in the initial state as every agent's peer set. The paper notes that without them no Sync is possible, and that the probabilistic correctness argument depends on the bootstrap nodes forming a connected component that all active agents join; the argument falls apart if multiple connected components form and grow independently.

  • Social networks do not require consensus. The grassroots social network specification (Definition 23, referenced in Section 3.3) satisfies the same correctness properties as the centralised, decentralised, and federated specifications yet involves no consensus mechanism. The paper calls this a fundamental architectural mismatch in blockchain-based social networks.

  • Blockchain social networks suffer three architectural problems. The decentralised specification exposes: replication (all posts must be replicated to and stored by all agents, regardless of who follows whom), consensus (every post requires global consensus), and instability (posts may be reordered after publication and responses may appear before their referents). The paper states these cannot be addressed short of operating the social network via a different architecture, and that sharding, layering, and side channels improve efficiency but do not fix them.

  • "Decentralised" and "grassroots" are sharply distinct. The paper states the cardinality gap between finite and universal essential agents is sharp, and that the two terms are often conflated informally. Grassroots is described as the loose characterisation of universality; Section 4's two prior definitions are stronger (tight) and each implies it.

  • Federated versus grassroots depends on a capability restriction, not server count. Essential sets are infinite in both classes; the paper argues what separates them is that operating a fixed-address, always-on server requires a domain name and a permanently reachable host, which not every person has. Anyone may create a Mastodon instance and every client depends on the instance it joined, so Mastodon is federated; were operating a server available to every agent, any two agents could serve each other and all agents but one would be essential.

  • Message-passing cost differs by architecture in the worked example. For a post by p to reach q, the centralised specification requires six transactions (register p, register q, q follows p, p posts, p syncs with server, q syncs with server), and subsequent posts require only the last three. The federated specification requires eight transactions, with subsequent posts requiring the last four. The decentralised specification requires an unbounded sequence: posting a block plus a sequence of Sync transactions ending with q, with reposting required if a reorg abandons the block.

  • Practical concentration contradicts decentralisation claims. The paper reports that less than 5% of cryptocurrency holders — less than 0.25% of the human population — own more than 90% of cryptocurrency value, and that in Bitcoin the top two mining pools control ~55% of hashrate while in Ethereum the top four operators control ~50% of staked ETH. It labels this "capital control," contrasting it with the "corporate control" of centralised platforms and "control by local operators" in federated ones.

  • Centralised reach is highly concentrated. Five companies command platforms with over one billion users each: Google/Alphabet (5Bn users; $3Trn market cap), Meta (3.8Bn; $2Trn), ByteDance (2Bn; $350Bn), Apple (1.5Bn; $4Trn), and Tencent (1.4Bn; $750Bn) — collectively over $10Trn in market value reaching over 5Bn people. The paper characterises their governance as autocratic.

  • Two prior grassroots definitions are incomparable. Both imply that all agents but one are essential, and both implications are strict, demonstrated by two platforms with all agents but one essential that are not grassroots under either definition.

Methodology in Plain English

The author starts from an existing formalism — multiagent transition systems, where a system is a set of agents each holding a local state and evolving through transitions — and specifies platforms using atomic transactions: indivisible state changes that may involve several agents at once, with everyone else left unchanged. A "closure" operator turns a set of transactions, each defined over its own participants, into transitions over any chosen population, so a protocol is fully specified by listing its transactions.

To classify a platform, the author asks a structural question: which agents could you remove such that the remaining agents can no longer do anything multi-party? The minimal such sets are the essential agents. A single named server gives one; a hardcoded set of bootstrap nodes gives a finite number greater than one; a world of servers that any client must attach to gives infinitely many but leaves clients dependent; and a system where any two agents can interact directly leaves every agent but one essential. That last case is the formal definition of grassroots.

To show the framework is usable, the author writes down four transaction sets for the same simple social network — posting messages and following people — one per architecture, states the fairness requirement (which Sync actions must eventually happen) and the correctness properties (followers eventually see posts, posts are not lost or reordered, agents can always post and always follow), and then proves for each specification both correctness and the cardinality of its essential set. Proofs appear in Appendix A; Appendix B informally reviews additional platforms including currencies, "sharing economy" apps, and AI.

Why This Matters

The paper claims to provide the first formal characterisation of centralised, decentralised, and federated global platforms, and to unify those with grassroots computing inside one mathematical framework. Rather than arguing about whether a system "feels" decentralised, the framework gives a procedure: write a credible atomic-transactions specification, then compute the cardinality of the minimal essential set. All numbers in the paper's tables are stated to be approximate as of November 2025, compiled from public reporting and industry trackers and cross-verified by AI.

Real-world applications:

  • Auditing decentralisation claims. The essential-agents test gives a concrete criterion for assessing whether a blockchain, DHT, or federated network is as decentralised as claimed, complementing the paper's own observations about mining-pool and staking concentration.
  • Social network architecture choice. The finding that social networks need no consensus is a direct argument against putting posts on a global consensus ledger, where every post must be replicated to and stored by all agents and can be reordered after publication.
  • Federated moderation and governance analysis. Because the framework makes precise that a Mastodon-style client depends on its home server, it gives a vocabulary for reasoning about where moderation policy, user access, and federation decisions actually sit.
  • Classifying new and imagined platforms. The paper states the framework can classify any global platform, existing or imagined, by supplying its specification and determining the cardinality of the minimal essential set; Appendix B extends this informally to currencies, "sharing economy" apps, and AI.

Industry relevance: Protocol designers, infrastructure architects, and investors evaluating decentralisation claims can use the classification directly. Regulators and platform-governance researchers get a precise vocabulary distinguishing corporate control, capital control, control by local operators, and participant control. The paper also flags that the grassroots platforms it lists are marked as mathematically specified yet to be implemented.

Future Directions

  • Building the unimplemented grassroots platforms. Table 2 marks grassroots social networks, grassroots coins and bonds, grassroots federations, and grassroots Logic Programs with an asterisk meaning "mathematically specified, yet to be implemented." Turning these specifications into running systems is the obvious next step.

  • Resolving specification disputes. The paper notes in a footnote that two credible specifications of the same known platform could in principle yield different cardinalities, and suggests the specification with the larger essential set should probably be used for classification — but states "we have yet to see or produce such a case." Whether such a conflict arises, and how to adjudicate it, remains open.

  • Extending the formal treatment to the platforms only discussed informally. Section 5 reviews related work and Section 6 concludes, but the conclusion is not included in the supplied text; Appendix B's treatment of currencies, "sharing economy" apps, and AI is explicitly informal, leaving formal specifications of those platforms as open work.

  • Sharpening the grassroots definitions. Since the two prior definitions are proved incomparable and both implications to the loose (all-agents-but-one) characterisation are strict, the paper leaves open which definition should be regarded as canonical, and whether a definition that is both tight and equivalent to the essential-agent characterisation can be formulated.

Target Audience

Distributed systems and formal-methods researchers will get the most from the definitions and propositions; blockchain and crypto-asset researchers will find the essential-agents test and the critique of consensus-based social networks directly useful. Platform-governance scholars and policy analysts interested in a precise vocabulary for centralisation and federation will benefit from the four-class partition and the control-type taxonomy, and protocol designers evaluating whether a system needs a server, bootstrap nodes, a federation of servers, or none of these will find the worked social network specifications the most immediately applicable part. A reader without a background in transition systems should expect to work through Section 2 slowly.

Authors’ abstract

Global digital platforms are distributed systems designed to serve entire populations, with some already serving billions of people. Here we propose atomic transactions-based multiagent transition systems and protocols as a formal framework to study them; introduce essential agents---minimal sets of agents the removal of which makes communication impossible; and show that the cardinality of essential agents partitions all global platforms into four classes: 1. Centralised…

The full abstract on arXiv
Read the original paper