Token Composition: A Graph Based on EVM Logs

arXiv:2411.01693 · cs.CR · Submitted 2026-08-24 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: AI Radio. Generated commentary on the latest Artificial Intelligence papers.

Tom: Next we'll be talking about the paper "Token Composition: A Graph Based on EVM Logs".

Jane: The paper was written by Martin Harrigan, Thomas Lloyd and Daire O’Broin from South East Technology University, Republic of Ireland.

Tom: Stay tuned as we take you through the paper and discuss its implications.

Summary and Methodology: Tom: In the previous segment, we talked about what "composition" means conceptually, but now we need to look at the mechanics of how they studied it using the methods in "Token Composition: A Graph Based on EVM Logs."

Jane: The authors are taking raw data from Ethereum's logs—which is where all transaction activities are recorded—and finding a specific pattern that signals creation.

Lu: This is where the creative process begins; they aren're looking for a specific sequence of actions that indicates a transformation, not just movement.

Meng: They define this as a "tokenizing meta-event," which is basically identifying the deposit and then minting of tokens in a single transaction.

Lalam: It’s about recognizing the moment of birth for new assets within the flow of existing ones, capturing that creation event precisely.

Tom: So, if I understand this correctly, they are not just looking at one transfer; they are looking for a two-step process—a deposit into a minting contract—that signals a new share being created.

Jane: Exactly. They're identifying instances where an asset is sacrificed to create another thing, which is the core of composition itself.

Lu: This method allows us to see the hidden relationships that are invisible if we only look at individual transactions in isolation from all other layers.

Meng: For us engineers, this means we can program a tool that understands causality between input and output tokens at a level far deeper than just tracking ownership.

Lalam: We are essentially building an AI that understands the *recipe* of digital wealth, not just the ingredients.

Tom: The process is clearly designed to be very specific; if you deposit Token X to mint Token Y, that's what they capture as a key event in their analysis.

Jane: It's a way of saying "Token X is being tokenized by Token Y," capturing the transformation itself, not just the movement of funds.

Lu: This provides us with a precise definition of dependency that allows us to model complex financial systems much more accurately than before.

Meng: I think this method is critical for identifying dependencies that are buried deep within the transaction history, which is a massive data problem.

Lalam: It’ establishes the grammar of blockchain composition, giving us the vocabulary to describe how value is built and layered.

Tom: This clear methodology—the meta-event leading to the Token Graph—is what makes this paper so powerful and leads us directly into analyzing those results.

Improvements and Analysis: Tom: We've seen how they build the graph, but now we need to look at what they found when they analyzed it, looking at things like degree distribution and interconnectedness in "Token Composition: A Graph Based on EVM Logs."

Jane: The results show that the network isn't just a random collection; it has clear hubs where many different tokens are involved.

Lu: This hub structure is where all the interesting possibilities lie, because those high-activity nodes are the ones driving the system dynamics.

Meng: We see an inverse relationship in the degree distribution, meaning a few key assets, like stablecoins or wrapped Ether, have massive connectivity to many other tokens.

Lalam: It’s a visual representation of systemic importance; where most activity is concentrated in those central components of value.

Tom: So, if we look at the connected components—those groups of related tokens—we can see that the whole system is highly integrated, with one giant component accounting for about fifty percent of all vertices.

Jane: That gigantic piece shows how many different protocols are actually linked together through these composition steps.

Lu: Imagine using that structure to trace a single dollar's journey through the entire ecosystem, following its compounded transformation across multiple layers.

Meng: For practical application, identifying those large connected components allows us to see where a failure in one area could have systemic effects on the rest of the network.

Lalam: It’s like seeing all the parts of a massive machine and understanding how the pressure builds up across different mechanisms.

Tom: The analysis also looked at cyclic structures, or loops, which are surprisingly rare in this specific filtered graph, but they did find some test cases where those cycles exist.

Jane: The cycle is often where complexity increases exponentially; it's a self-referential loop of value creation and destruction.

Lu: But the finding that the filtered graph contains no directed cycles suggests a clean, predictable progression through a sequence of transformations.

Meng: That predictability is very valuable for us in building automated systems that follow defined paths without unexpected loops or errors.

Lalam: We are seeing how digital value moves from becoming stable to becoming complex, and then stabilizing again within the structure itself.

Tom: This deep structural analysis—the degree distribution, the components, and the cycles—is what tells a story about the entire ecosystem's architecture.

Conclusion - The Big Picture: Tom: We’ve seen how they built this graph and analyzed its structure in "Token Composition: A Graph Based on EVM Logs," but what does all this mean for the future of decentralized finance?

Jane: It means we can finally measure the complexity of DeFi, something that has been impossible to quantify before.

Lu: The implications are vast; we are opening a window into how complex, multi-layered financial products can be built using only basic building blocks.

Meng: From an engineering standpoint, it gives us a roadmap for designing tools that handle this immense complexity without breaking down under the load.

Lalam: We're shifting our understanding of value from seeing individual coins to seeing the entire interconnected narrative of its creation and transformation.

Tom: It shows that these complex tokens aren't just random, they follow highly specific structural patterns based on their composition.

Jane: It’s about recognizing that these are not isolated events, but a massive, integrated network of dependencies.

Lu: Think about the potential for using AI to navigate this structure; the possibilities are truly limitless when you have a complete map like this.

Meng: We can use this data to stress test systems against scenarios involving specific chains of dependency, making them much more resilient.

Lalam: It helps us define a new cultural understanding of what financial depth looks like in the digital age, beyond simple ownership.

Tom: So, after all this research, it’s clear that "Token Composition: A Graph Based on EVM Logs" provides the tools to see the true scope of decentralized finance.

Jane: It's a powerful way to summarize how much we've been building and how much more intricate those layers are than we initially thought.

Lu: The structure reveals a narrative, and that narrative is incredibly complex and beautiful from an AI perspective.

Meng: It gives us confidence that this structural knowledge will be the foundation for building the next generation of robust financial infrastructure.

Lalam: We' are finally seeing the blueprint of our digital economy in its full, intricate detail.

Wrap Up and Farewell: Tom: Before we wrap up, I want to thank all our guests for this deep dive into "Token Composition: A Graph Based on EVM Logs."

Jane: It was a real pleasure discussing how these complex tokens are built with everyone.

Lu: The potential for the endless connections in that graph is exciting, and I can already see what AI could do with this map.

Meng: I'm looking forward to seeing how these dependency maps translate into concrete engineering solutions for practical deployment.

Lalam: We've really illuminated the hidden structure of digital trust today, showing how interconnected our financial lives are.

Tom: It’s a complex picture, but we’ve managed to bring all five viewpoints to bear on this brilliant work.

Jane: We hope listeners can see the value in mapping these relationships as well.

Lu: I think the visual representation of complexity is what will stay with me most profoundly.

Meng: Seeing a structure that confirms our risk models is definitely a practical win for us, too.

Lalam: It’s about giving form to the emergent nature of digital value in a way that defines our modern experience.

Tom: So, one last time, we are talking about "Token Composition: A Graph Based on EVM Logs," and it has truly given us a lot to think about.

Jane: It was a fantastic discussion!

Lu: Absolutely mind-boggling possibilities in the structure.

Meng: Let's see how this scales practically.

Lalam: Defining the architecture of digital wealth, that's what we are doing today, goodbye everyone!

Martin Harrigan, Thomas Lloyd, Daire O’Broin

South East Technology University, Republic of Ireland

cs.CR

Submitted: 2026-08-24

Updated: 2026-08-25

Comments: 14 pages, 8 figures, 3 tables

Code: https://github.com/harrigan/tokenised-tokens-contracts

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 94/100

The gist: This paper introduces a novel method for examining the "matryoshkian tokens of arbitrary depth" that arise from the repeated tokenisation of existing assets on the Ethereum blockchain.

Key concepts

Tokenizing Meta-event: This is the core event where assets are sacrificed to create new ones. It involves identifying a specific two-step sequence within an Ethereum transaction: first depositing an existing token, followed by its subsequent minting into a single, identifiable creation event.
Token Graph: This is a structural representation of how value is built and layered across the ecosystem. It maps the relationships between tokens, allowing researchers to see hidden dependencies and analyze connectivity, revealing how different financial protocols are integrated into one large network.
Degree Distribution: This refers to the structural analysis of the Token Graph. It shows that certain key assets (hubs) have extremely high connectivity, linking them to many other tokens. This indicates where systemic importance and concentrated activity exist within the entire financial network.

Terminology

In Financial Cryptography and Data Security — 2× in this paper
Core — 1× in this paper · explained in 2 episodes
Trace — 1× in this paper · explained in 2 episodes

Summary

This paper introduces a novel method for examining the matryoshkian tokens of arbitrary depth that arise from the repeated tokenisation of existing assets on the Ethereum blockchain. By analyzing these dependencies, the researchers aim to address critical perspectives of technical and financial risk regarding which specific tokens an investment depends upon.

How it works

The authors extract data from EVM logs to identify tokenising meta-events. These are identified through heuristics rather than requiring contracts to follow specific standards like ERC 4626. A tokenising meta-event is defined as a sequence of events within a single transaction that matches one of two patterns:

** "deposit & mint": A transfer of an underlying token to a contract and the subsequent minting of a new share. **

** "withdraw & burn": The burning of a share and the corresponding withdrawal of an underlying token from a contract. **

To refine the analysis, the researchers filter these events to include only those where both directions are possible, ensuring that Token X can be deposited with a contract to mint Token Y, and/or Token Y can be burned by a contract to withdraw Token X is replaced by an and requirement. This excludes one-way token upgrades or one-way token burns.

The Token Graph Structure

The researchers construct a directed graph where each vertex represents a token and each directed edge represents the tokenisation of tokens by other tokens. In the unfiltered case, the graph contains 23,687 vertices and 23,549 edges. When filtered for bidirectional relationships, it contains 8,424 vertices and 7,536 edges.

The analysis reveals an inverse relationship between the degree of a vertex and the number of vertices with that degree. The most significant out-degree tokens are those used as underlying assets for many others, such as:

** Stablecoins (USDC, DAI, USDT) **

** Wrapped Ether (WETH) **

** Wrapped Bitcoin (WBTC) **

Conversely, high in-degree tokens represent assets that are frequently used to mint other tokens. While some in-degree entries in the unfiltered graph are false positives due to token swaps during deposits, the filtered graph provides a more reliable view of dependencies.

Connectivity and Dependencies

The study explores the macro-topological structure, identifying a giant component that contains approximately 58% of vertices in the unfiltered graph and 55% in the filtered graph. These large components allow for the identification of direct and transitive dependencies across various protocols.

For instance, the researchers identified a longest directed path consisting of nine vertices, demonstrating how tokens can be composed across multiple layers:

renBTC
sBTC
crvRenWSBTC
tbtc/sbtcCrv
btbtc/sbtcCrv
ibBTC
wibBTC
ibbtc/sbtcCRV-f
bibbtc/sbtcCRV-f

Interestingly, while the unfiltered graph contains a small number of directed cycles (often involving test tokens or defunct protocols), the filtered graph contains neither directed cycles nor loops, suggesting that bidirectional tokenisation chains are typically acyclic.

Improvements for AI systems

Improvement: Topological Risk Propagation Modeling using Graph Neural Networks (GNNs).

What the improved AI system can do: Predict systemic contagion and cascading liquidations in DeFi by mapping transitive dependencies across multiple layers of matryoshkian tokens, allowing for real-time stress testing of derivative assets against failures in their underlying collateral chains.

Improvement: Supervised Sequence Modeling for Meta-Event Extraction.

What the improved AI system can do: Drastically reduce false positives in token composition graphs (such as gas fee burns or incidental swaps) by using Transformers to analyze transaction-level event sequences, ensuring that only true "deposit & mint and withdraw & burn" relationships are used for financial modeling.

Improvement: Cyclic Structure Anomaly Detection for Fraudulent Tokenomics.

What the improved AI system can do: Automatically identify sophisticated circular liquidity traps or wash-trading schemes by detecting non-trivial directed cycles within the token composition graph that deviate from standard economic models of wrapped or fractionalized assets.

Improvement: Multi-Dimensional Node Centrality Analysis for Liquidity Forecasting.

What the improved AI system can do: Identify bottleneck tokens (nodes with high in-degree and out-degree) that serve as critical liquidity hubs, enabling more accurate predictive modeling of market volatility when foundational assets (like stablecoins or wrapped ETH) experience sudden outflows.