Token Snowball: Recursive Tokens & Avalanche Consensus

• Token Snowball is a dual-faceted phenomenon describing recursive token wrapping on Ethereum and a stake-weighted consensus in Avalanche.
• Empirical analysis shows token composition can reach depths of up to 9 layers with heavy-tailed degree distributions and critical hub tokens.
• The Avalanche mechanism uses stake-weighted sampling to balance liveness and safety, though deep compositions may accumulate systemic risks.

Token Snowball denotes two distinct but technically rich phenomena within blockchain systems: (1) the recursive token composition dynamics that generate arbitrarily deep hierarchies of “wrapped” assets within the Ethereum ecosystem, yielding what is called the token-snowball or “matryoshkian” effect; and (2) the stake-weighted Snowball consensus protocol underpinning Avalanche, where voting power accrues snowball-like via repeated stake-weighted sampling. Both embody growth and amplification via cumulative, recursive processes, but with substantially different mathematical and systemic implications.

1. Token-Composition Graphs and the Matryoshkian Effect

Token snowballing on Ethereum arises via recursive token composition: tokens that represent, wrap, fractionalize, or otherwise synthesize other tokens, with protocols chaining these operations arbitrarily deeply. Formally, let $G=(V,E)$ denote the directed “token-composition graph,” with vertices $V$ indexing all ERC-20 and ERC-721 token contracts observed to participate in at least one tokenising meta-event (i.e., minting or burning as a function of underlying token deposits or withdrawals). A directed edge $(u \to v) \in E$ records the on-chain observation of token $v$ being minted from $u$ (deposit-wrapping) as well as its inverse (burn-to-redeem).

Empirical results covering the Ethereum blockchain up to block 16,685,101 censor the graph $G$ as follows:

Graph	$|V|$	$|E|$	Filtering Criteria
Unfiltered	23,687	23,549	Edge observed in either direction
Filtered	8,424	7,536	Both deposit/mint and withdraw/burn observed

This graph framework encodes both the combinatorial complexity and the practical chain of custody linking complex DeFi assets to their atomic underlyings (Harrigan et al., 2024).

2. Matryoshkian Depth: Formal Definition and Quantification

The depth of token composition captures the length of the longest chain of recursive wrapping. Define for each $t \in V$:

• $$\operatorname{In}(t) = \{u \in V : (u \to t) \in E\}$$ (immediate underlyings)
• Base tokens $B = \{t \in V : \operatorname{In}(t) = \emptyset\}$

The matryoshkian depth $d:V\to \mathbb{N}$ is

$$d(t) = \begin{cases} 0, & t\in B \ \max_{u\in \operatorname{In}(t)} [ d(u) + 1 ], & \text{otherwise} \end{cases}$$

Computationally, $d(t)$ is the length of the longest directed path ending at $t$ originating from a base token. Topological traversal efficiently computes $d(t)$ for all $t$ in acyclic graphs.

Empirical findings show:

• Maximum observed depth $d_{\max}=8$ (9 tokens in chain), e.g., renBTC $\rightarrow$ sBTC $\rightarrow$ crvRenWBTC $\rightarrow$ tbtc/sbtcCrv $\rightarrow$ btbtc/sbtcCrv $\rightarrow$ ibBTC $\rightarrow$ wibBTC $\rightarrow$ ibbtc/sbtcCRV-f $\rightarrow$ bibbtc/sbtcCrv-f.
• Distribution: $\sim$70% tokens at depth 0; $\sim$20% at depth 1; $\sim$8% at depths $2-5$; $<1\%$ at depth $\geq 6$.

3. Graph-Theoretic Structure and Topology

The graph $G=(V,E)$ exhibits nontrivial graph-theoretic features:

• Degree Distribution: Both in-degree and out-degree feature heavy-tailed, near power-law profiles—central nodes (e.g., USDC, DAI, WETH) act as hubs with numerous incident edges, while most tokens participate in only one- or two-step compositions.
• Connected Components: In the unfiltered graph, the giant weakly connected component spans $|V|=13,794$ ($\sim 58\%$ of tokens), containing intertwined DeFi primitives (stablecoins, vaults, synthetic assets). Filtered graphs remain dominated by a giant component, but both filtered and unfiltered graphs comprise many isolated two-token wrappers and small subsystems.
• SCCs and Cycles: The filtered graph is acyclic (no strongly connected components with more than one vertex), whereas the unfiltered graph contains a handful of short cycles among inactive or test tokens. This reflects unidirectionality in most asset-wrapping deployments, with rare edge cases in testing environments.

These topological signatures map the system-wide risk and dependency structure induced by deep token compositions (Harrigan et al., 2024).

4. Empirical Methodology for Token Snowball Discovery

The empirical approach for constructing $G$ leverages on-chain event logs:

1. Data Collection: Use Ethereum eth_getLogs to retrieve all ERC-20 Transfer events from genesis to the specified block.
2. Transaction Grouping: Events aggregated by transaction hash.
3. Tokenising Meta-event Detection: Within each transaction, a deposit (transfer from user to contract) coincides with a mint (issue of new token), and reciprocally a burn event triggers a withdrawal (contract to user). The meta-event $(u, v)$ is labeled whenever both directions are observed.
4. Bidirectionality Filtering: Retain only $(u, v)$ pairs with at least one deposit-mint and one burn-withdraw observed, guaranteeing full composability and redeemability.
5. Graph Construction: Vertices $V$ consist of all tokens in filtered pairs; $E$ consists of directed edges from underlyings to their corresponding shares.

Pseudocode and processing sequence are explicitly detailed in (Harrigan et al., 2024).

5. Systemic and Security Implications

The layered, directed structure of token snowballing exposes several nontrivial implications:

• Prevalence of Non-Atomic Assets: A majority of tokens in active economic use are not primitive, but aggregates recursively defined through extensive cross-protocol wrapping. Over half of all filtered tokens participate in the primary component linking stablecoins, LP tokens, yield vaults, and synthetic shares.
• Accumulated Risk: Each recursive layer accrues additional counterparty, liquidity, oracle, and contract risk. While observed matryoshkian depths seldom exceed 8, there are no protocol-level restrictions precluding significantly deeper chains, particularly from composable index- or staking-protocols.
• Opacity and Fragility: Deeply nested asset compositions can obscure actual exposure, causing surface-level tokens to mask complex risk positions several layers down.
• Risk Monitoring Capability: The token-composition graph $G$ makes explicit the deep dependency structure. Analytical tools leveraging $d(t)$, component topology, and degree statistics may inform collateral eligibility, system-wide risk assessments, and protocol-level security audits.

Such rigorous mapping translates the intuitive metaphor of “Russian dolls” into quantifiable systemic insight (Harrigan et al., 2024).

6. Stake-Weighted Snowball Protocol in Avalanche

“Token Snowball” also designates the stake-weighted variant of the Snowball consensus protocol in Avalanche (Kniep et al., 2024):

• Protocol Mechanics: Each validator $v$ has stake $\mathsf{stake}(v)$, total stake $n = \sum_v \mathsf{stake}(v)$. In each synchronous round, every node samples $k$ peers (weighted by stake), and assigns a “chit” to the majority color if at least $\alpha$ of $k$ agree. When a color accrues $\beta$ consecutive chits, the node finalizes that preference.
• Parameterization: Typical Avalanche settings are $k=20, \alpha=15, \beta=20$.
• Adversary Model: For subset of adversarial stake $f$, define $\alpha_{\mathsf{adv}} = f/n$. Two adversarial regimes: naïve (only sees queried node votes), and informed (sees fraction of honest color preferences each round).

Resilience properties are as follows:

• Liveness Vulnerability: An adversary with $f = \Omega(\sqrt{n})$ stake ($\approx$ 5.2% for $n=2000$, dropping to 2.8% if adversary is informed) can indefinitely stall protocol termination.
• Safety Breaks: Probability an adversary can cause finalization divergence grows exponentially with its stake fraction and decreases linearly with network size (e.g., in networks of $n=3000$ with $25\%$ adversarial stake, safety is compromised in approximately 265 rounds on expectation).
• Implication: Token Snowball, as deployed in Avalanche, guarantees only Byzantine reliable broadcast (agreement, validity, integrity), but cannot guarantee termination (liveness) or strong finality under adversarial conditions with nontrivial stake concentration.

The token-snowball mechanism of Avalanche, therefore, exhibits a liveness/safety tradeoff profile fundamentally distinct from classical PoS BFT, with much tighter adversarial bounds (Kniep et al., 2024).

7. Visualizations and Canonical Examples

Several key visual representations elucidate token snowball structure:

• Composition Graphs: Toy $n$-vertex graphs with multiple wrapping pathways (one-way and bidirectional), highlighting two-token cycles and multi-edge triangles.
• Protocol Subgraphs: Isolated subgraphs exemplify specific DeFi protocols—e.g., ~12-vertex Angle protocol subgraph showing governance token staking and LPing, and ~30-vertex JPEG’d protocol component tracing NFT-collateralized synthetic ETH through LPs, gauges, xTokens, and their recursive wrappers.
• Degree Distributions: Log-log plots exhibit heavy-tailed (near power-law) in- and out-degree histograms for both filtered and unfiltered graphs.

Canonical chain: renBTC $\to$ sBTC $\to$ crvRenWBTC $\to$ tbtc/sbtcCrv $\to$ btbtc/sbtcCrv $\to$ ibBTC $\to$ wibBTC $\to$ ibbtc/sbtcCRV-f $\to$ bibbtc/sbtcCrv-f, demonstrates the potential for protracted snowballing in contemporary DeFi tokens (Harrigan et al., 2024).

---

References:

• "Token Composition: A Graph Based on EVM Logs" (Harrigan et al., 2024)
• "Quantifying Liveness and Safety of Avalanche's Snowball" (Kniep et al., 2024)