Fairness in Token Delegation: Mitigating Voting Power Concentration in DAOs (2510.05830v1)
Abstract: Decentralized Autonomous Organizations (DAOs) aim to enable participatory governance, but in practice face challenges of voter apathy, concentration of voting power, and misaligned delegation. Existing delegation mechanisms often reinforce visibility biases, where a small set of highly ranked delegates accumulate disproportionate influence regardless of their alignment with the broader community. In this paper, we conduct an empirical study of delegation in DAO governance, combining on-chain data from five major protocols with off-chain discussions from 14 DAO forums. We develop a methodology to link forum participants to on-chain addresses, extract governance interests using LLMs, and compare these interests against delegates' historical behavior. Our analysis reveals that delegations are frequently misaligned with token holders' expressed priorities and that current ranking-based interfaces exacerbate power concentration. We argue that incorporating interest alignment into delegation processes could mitigate these imbalances and improve the representativeness of DAO decision-making.
Summary
- The paper presents a comprehensive analysis revealing that current delegation mechanisms misalign token holders’ interests with delegate behavior, fueling voting power concentration in DAOs.
- It employs a multi-modal methodology that combines on-chain data with off-chain forum insights and uses metrics like Gini and Nakamoto coefficients to quantify centralization.
- The study advocates for redesigning delegation interfaces to prioritize ideological alignment, thereby mitigating governance capture and promoting fairer decision-making.
Fairness in Token Delegation: Mitigating Voting Power Concentration in DAOs
Introduction and Motivation
This paper presents a comprehensive empirical analysis of token delegation mechanisms in Decentralized Autonomous Organizations (DAOs), focusing on the systemic concentration of voting power and the misalignment between token holders' interests and delegate behavior. The authors combine on-chain data from five major protocols (Uniswap, Aave, Arbitrum, ENS, Compound) with off-chain governance forum discussions from 14 DAOs, constructing a multi-modal dataset to quantify ideological alignment and power distribution. The paper is motivated by persistent issues in DAO governance: voter apathy, interface-driven visibility biases, and the rich-get-richer dynamics that undermine decentralization and representative decision-making.
Figure 1: Delegation interface on Tally, where delegates are ranked by voting power, reinforcing visibility and centralization.
Data Collection and Methodology
The analysis leverages full Ethereum archival data and extensive forum scraping, linking user identities across platforms via ENS and Tally profiles. The pipeline (Figure 2) integrates on-chain governance actions (proposal creation, vote casting, delegation changes) with off-chain discussion signals, using LLM-based keyword extraction and sentiment analysis to build behavioral profiles.
Figure 2: Overview of the interest-aligned delegation pipeline, mapping forum activity to on-chain addresses and extracting ideological profiles for alignment analysis.
Identity resolution is performed with a multi-tiered confidence system, prioritizing exact ENS matches and manual verification to ensure reliable linkage between forum users and wallet addresses. The resulting dataset enables granular analysis of both structural and ideological aspects of DAO governance.
On-Chain Analysis: Concentration and Network Structure
The paper quantifies concentration using Gini and Nakamoto coefficients, revealing extreme inequality in both token holdings and delegated voting power. For instance, Gini coefficients for voting power exceed 0.94 across all protocols, with some DAOs (ENS, Uniswap, Compound) reaching 0.99, indicating near-complete centralization. The Nakamoto coefficient analysis shows that as few as 1–8 addresses can control one-third to one-half of the token supply in major DAOs.
Figure 3: CDF of top holders, illustrating the extreme concentration of token ownership in major DAOs.
Delegation networks are modeled as directed graphs, exhibiting sparse, fragmented, and disassortative hub-and-spoke topologies. Most delegators assign their entire balance to a single delegate, further amplifying centralization. The largest weakly connected components contain only a small fraction of nodes, and degree assortativity is consistently negative, confirming the dominance of a few high-degree delegates.
Off-Chain Analysis: Proposal Categorization and Voter Interest Discovery
The authors develop a unified taxonomy for proposal categorization, integrating forum-specific and Messari categories into 11 high-level groups and 48 subcategories. Proposal importance is assigned using GPT-5, with prompts designed to capture both category and significance.
Figure 4: Prompt used for Proposal Categorization, operationalizing the taxonomy for LLM-based labeling.
Figure 5: Sankey diagram of proposal categorization, showing the distribution of topics and importance across the analyzed dataset.
Voter interests are extracted from forum posts using LLM-based keyword extraction, with comparative evaluation of GPT-5 variants and embedding-based models. The gpt-5-mini model is selected for its balance between specificity and generality. Keyword sets are aggregated at the voter level and embedded using all-MiniLM-L12-v2 for semantic clustering.
Figure 6: Example of a proposal in the governance forum, illustrating the joint analysis of root and response posts for interest extraction.
Figure 7: Prompt used for Keyword Extraction, guiding LLMs to produce interpretable and granular interest signals.
Hierarchical clustering (Ward’s method) yields five distinct voter clusters, visualized via dendrogram and t-SNE projection.
Figure 8: Dendrogram from hierarchical clustering, identifying five major voter interest groups.
Figure 9: t-SNE plot visualizing cluster assignments, highlighting the semantic separation of voter interests.
Cluster characterization via word clouds reveals distinct ideological profiles: Finance-Driven, Governance-Driven, Arbitrum DeFi-Driven, ENS Ecosystem-Driven, and Innovation-Driven voters.
Figure 10: Word clouds of top keywords for each voter cluster, summarizing the dominant interests and priorities.
Empirical Findings and Numerical Results
The analysis demonstrates that delegation mechanisms, as currently implemented, exacerbate power concentration and ideological misalignment:
- Delegation is frequently misaligned: Many token holders delegate to highly visible actors whose voting behavior does not reflect their stated interests.
- Ranking-based interfaces reinforce centralization: Default sorting by voting power on platforms like Tally drives further accumulation of influence among top delegates.
- Participation is broad but ineffective: While many users act as delegators, their choices are funneled toward a small elite, with diversification across multiple delegates being rare.
- Strong numerical results: Gini coefficients for voting power and token holdings consistently exceed 0.94; Nakamoto coefficients indicate that a handful of addresses can control a majority stake in governance.
Implications and Future Directions
The findings challenge the assumption that delegation naturally improves representativeness in DAOs. The authors argue for the integration of interest alignment into delegation interfaces, enabling token holders to select delegates based on shared values and historical behavior rather than visibility or accumulated power. This approach could mitigate systemic vulnerabilities such as governance capture and ossification.
Practically, the methodology and dataset provide a foundation for designing fairness-aware delegation mechanisms and for auditing the representativeness of DAO governance. The results have direct relevance for platforms like Tally, which could implement alignment-aware sorting and recommendation systems to redistribute delegator support more equitably.
Theoretically, the paper highlights the necessity of multi-modal analysis in decentralized governance, combining structural, behavioral, and ideological data to fully understand decision-making dynamics. Future research may explore privacy-preserving delegation protocols, artificial delegates for quota compliance, and cost-minimization strategies to further enhance fairness and accountability.
Conclusion
This paper delivers a rigorous, data-driven assessment of token delegation in DAOs, revealing that current mechanisms amplify power concentration and misalignment between token holders and delegates. By combining on-chain and off-chain data, and leveraging LLMs for interest extraction, the authors provide actionable insights for both researchers and practitioners. The work underscores the need for interest-aligned delegation processes and transparent interfaces to realize the promise of participatory, decentralized governance.
Paper Prompts
Sign up for free to create and run prompts on this paper using GPT-5.
Top Community Prompts
Explain it Like I'm 14
1) What this paper is about (overview)
This paper looks at how online groups called DAOs make decisions. In DAOs, people who own special tokens can vote on changes. Many people don’t have time to vote, so they “delegate” their votes to someone else. The paper shows that today’s delegation systems often give too much power to a few very visible delegates, even when those delegates don’t share the same priorities as the people who gave them votes. The authors paper this problem and suggest ways to make delegation fairer and more aligned with what token holders actually care about.
2) What the researchers wanted to find out (key questions)
- Are delegations in DAOs aligned with what token holders care about, based on what those holders say in discussion forums?
- Do current interfaces (like websites that list delegates) accidentally push people to pick the same high-profile delegates, concentrating power?
- Can we design better delegation methods that match voters with delegates who share their interests?
3) How they studied it (methods explained simply)
The researchers combined two kinds of information:
- On-chain data: This is like the public record of all the “official actions” on the blockchain—who owns tokens, who delegates to whom, when votes happen, and what proposals are created.
- Off-chain data: This is what people say in DAO discussion forums—posts where token holders talk about ideas, priorities, and opinions.
To connect the two:
- They linked forum usernames to blockchain wallet addresses using public profiles and .eth names (like online nicknames tied to wallets). Think of it like matching someone’s forum account to their student ID.
- They used AI LLMs to read forum posts and pull out topics and interests (for example, “treasury management” or “ecosystem security”). This is like having a smart assistant summarize lots of messages to figure out what each person cares about.
- They checked whether delegations matched these interests, and they measured power concentration using simple fairness metrics:
- Gini coefficient: A number between 0 and 1. Close to 1 means “very unequal” (a few have most of the power).
- Nakamoto coefficient: How many top actors it takes to control a big chunk (like half) of the power. A small number means power is very concentrated.
- They also built “delegation networks” (maps showing who delegates to whom) to see if they look like “hub-and-spoke” systems—many small delegators pointing to a few big delegates.
4) What they found and why it matters (main results)
- Power is extremely concentrated: A few delegates hold most of the voting power in several major DAOs. For some, the inequality score (Gini) is close to 1, which means it’s very unfairly distributed.
- “Rich get richer” interfaces: Websites that list delegates by how much power they already have make it more likely that new delegations go to the same big names. This snowballs their influence.
- Misalignment is common: Many token holders delegate to people who do not match their stated interests on forums. In other words, the person voting for them may not represent what they actually care about.
- Delegation networks look like hubs: Most delegators pick just one delegate, and a few “hub” delegates receive support from many accounts.
- Community interest groups exist: The authors found clusters of interests, such as:
- Finance-focused (treasury, accountability, performance)
- Governance-focused (rules, ownership, public goods funding)
- DeFi-focused (liquidity incentives, risk parameters—especially around Arbitrum)
- ENS-focused (technical upgrades, working groups, operations)
- Innovation-focused (new initiatives like NFTs and broader decentralization)
Why this matters: When decisions are made by a small, unrepresentative group, DAOs risk becoming less fair, easier to “capture” by powerful actors, and less responsive to what the community actually needs.
5) What this could change (implications and impact)
The paper suggests building delegation tools that:
- Match voters with delegates based on shared interests, not just popularity.
- Make alignment visible on delegation interfaces (so you can see who truly represents your values).
- Reduce “rich-get-richer” sorting by avoiding default rankings that favor already-powerful delegates.
If DAOs adopt these ideas, decisions could better reflect the whole community, power would be more balanced, and governance would be fairer and more resilient. The authors also plan to share their data and code so others can test and improve these methods, potentially influencing major platforms to redesign how people choose delegates.
Knowledge Gaps
Knowledge Gaps, Limitations, and Open Questions
Below is a focused list of what remains missing, uncertain, or unexplored in the paper. Each item is framed to be concrete and actionable for future research.
- Causal evidence on UI-driven concentration: The claim that ranking-based interfaces (e.g., default sorting by voting power on Tally) exacerbate concentration is not supported by causal analyses. Field experiments, A/B tests, or natural experiments (e.g., observing delegation flows before/after UI sorting changes) are needed to quantify the effect sizes and mechanisms.
- Incomplete on-chain coverage and protocol heterogeneity: The on-chain analysis excludes or underrepresents governance events for some protocols (e.g., missing VoteCast for Aave and Arbitrum, missing DelegateVotesChanged for AAVE), raising comparability issues due to heterogeneous contract designs and off-chain voting. A standardized cross-protocol event mapping and inclusion of off-chain voting platforms (e.g., Snapshot) are required.
- Cross-chain generalizability: Several analyzed governance forums belong to DAOs on non-EVM chains (e.g., Solana-based Wormhole, Jito), yet the data pipeline focuses on Ethereum and Arbitrum. Extending identity resolution and governance action collection to non-EVM ecosystems is needed to validate generalization.
- Sparse and potentially biased identity linkage: The final interest analysis is based on 391 forum–to-wallet matched addresses out of 86,445 Tally delegates, with only 284 high-confidence matches. A more robust, multi-signal entity resolution (e.g., ENS, GitHub/Twitter, signed statements, on-chain attestations) and a sensitivity analysis of false positives/negatives are needed to address sampling bias and external validity.
- Representativeness of voiced preferences: The approach infers token holders’ interests from forum participants, but most token holders are silent. Methods to elicit and model preferences of non-posting holders (e.g., surveys, preference elicitation tools, in-wallet questionnaires) are needed to avoid overfitting to vocal minorities.
- Unspecified and unvalidated “interest alignment” metric: The paper describes extracting interests and mentions comparing them to delegates’ historical behavior, but lacks a formal alignment score, aggregation rules, or weighting by token holdings and proposal importance. A transparent, reproducible alignment metric and validation against human judgments and voting outcomes are missing.
- LLM dependency without accuracy, robustness, or bias evaluation: Keyword extraction and categorization rely on GPT-5 family models without reporting precision/recall, inter-annotator agreement with human labels, prompt sensitivity, or cross-model robustness. Rigorous evaluation, error analysis, and bias audits are needed.
- Cluster stability and interpretability: Voter interest clusters are produced via embeddings and Ward’s method, but there is no stability assessment (e.g., bootstrap, silhouette scores), temporal robustness testing, or external validation. Linking clusters to concrete governance behaviors (proposal types, vote patterns) remains unexplored.
- Temporal dynamics and event studies: Delegation and interests change over time (e.g., around contentious proposals). The paper lacks longitudinal analyses (e.g., time series of concentration/interest alignment, event studies around major votes) that could reveal dynamics and causal structure.
- Outcome-level validation: The paper does not test whether interest-aligned delegations would materially change proposal outcomes, quorum attainment, or minority representation. Counterfactual simulations or replay studies that reroute delegations using proposed alignment rules are needed.
- Formal fairness objectives and metrics: “Fairness” is argued but not formalized. Clear objectives (e.g., quota compliance, Gini influence reduction, representation parity across interest clusters) and trade-off analysis (efficiency vs. representativeness) are necessary to design and evaluate interventions.
- Design and evaluation of interest-aware interfaces: The pipeline suggests interest-aligned recommendations but omits concrete design choices (ranking/scoring functions, explanations, UI affordances), user testing, and measurement of unintended consequences (e.g., echo chambers, fragmentation).
- Strategic manipulation and adversarial robustness: Interest-based discovery is susceptible to gaming (astroturfing, sybil accounts, fabricated bios, coordinated posting). Threat modeling and defenses (proof-of-personhood, anti-sybil mechanisms, reputation systems, audit trails) are unaddressed.
- Delegate behavior modeling: The paper does not systematically evaluate whether delegates’ votes and proposal sponsorship actually reflect their stated interests/bios over time. Methods to detect inconsistency (e.g., divergence metrics between stated positions and vote records) are needed.
- Multi-delegate splitting and delegation design space: Although multi-delegate splitting is mentioned, the paper does not analyze its prevalence or impact on concentration and alignment. Experiments with splitting policies (caps, quotas, dynamic routing) could quantify benefits and risks.
- Weighting by stake and participation: Interest alignment and clustering ignore token-weighted influence, delegator sizes, or vote participation rates. Incorporating stake-weighted preferences and turnout adjustments is required to reflect actual governance impact.
- Normative and ethical considerations: Mapping forum identities to wallets via ENS raises privacy and consent issues not discussed (e.g., do users expect cross-platform linkage?). Ethical guidelines and data governance policies for such linkage are needed.
- Economic and bribery dynamics: The paper does not analyze how economic incentives (e.g., bribes, vote-buying, paid delegates) interact with delegation concentration and interest alignment. Integrating on-chain transfer patterns, bribe markets, or governance attack indicators could reveal drivers of misalignment.
- Cross-platform ecosystem coverage: Tally-centric analysis overlooks other delegation and governance portals. Broader platform inclusion and comparative UI studies are needed to assess generalizability and platform-specific biases.
- Reproducibility and transparency: Code and datasets are “planned” for release; without them, replication is limited, especially given stochastic LLM outputs. Full release of prompts, model versions, seeds, preprocessing scripts, and matched-identity lists (with appropriate privacy safeguards) is necessary.
Practical Applications
Overview
Below are practical applications derived from the paper’s findings, methods, and innovations, grouped into immediate and long-term horizons. Each item notes relevant sectors, potential tools/products/workflows, and assumptions or dependencies that affect feasibility.
Immediate Applications
- Bold: Interest‑aligned delegate discovery widgets — Sector: software (web3 UX), finance
- Description: Integrate alignment scores and issue filters into delegation interfaces (e.g., Tally) so token holders can sort and filter delegates by shared priorities rather than raw voting power. Provide “Why this delegate?” explanations grounded in forum topics and historical votes.
- Tools/Workflows: LLM-based topic extraction from forum posts; mapping delegates’ historical votes to categories; UI components for alignment filters; backend APIs serving alignment scores.
- Assumptions/Dependencies: Accurate identity linkage (forum ↔ ENS/Tally ↔ wallet); sufficient public forum data; acceptable LLM inference costs; user consent and privacy controls for off-chain/on-chain linkage.
- Bold: Governance “Wahl‑O‑Mat” for DAOs (issue questionnaire → delegate recommendation) — Sector: software (civic tech), finance
- Description: A short survey embedded in wallets/governance portals to capture token‑holder priorities, returning a ranked list of aligned delegates with transparency on overlaps and conflicts.
- Tools/Workflows: Lightweight questionnaire; interest‑to‑topic mapping; recommender service; periodic retraining as forums/votes evolve.
- Assumptions/Dependencies: Reliable topic taxonomy; ongoing ingestion of forum and on‑chain data; UI distribution via major platforms/wallets.
- Bold: Fairness‑aware ranking A/B tests — Sector: platform operations, software
- Description: Run controlled experiments on delegate list ordering (power‑based vs. alignment‑based vs. randomized blends) to measure impacts on concentration, turnout, and misalignment.
- Tools/Workflows: Experiment framework; telemetry for conversion/engagement; metrics dashboards (Gini, Nakamoto coefficient, misalignment rates).
- Assumptions/Dependencies: Platform cooperation (e.g., Tally); ethical review for user experiments; clear KPIs.
- Bold: Concentration and misalignment risk dashboards — Sector: finance (treasury/governance), analytics
- Description: Provide near‑real‑time monitoring of voting power concentration, delegation inflows, and delegate‑constituent alignment per protocol/proposal to flag governance capture risks.
- Tools/Workflows: Archive node or subgraph ingestion; computation of Gini/Nakamoto; misalignment heatmaps; alerting.
- Assumptions/Dependencies: Stable on‑chain data pipelines; reproducibility and versioning; community buy‑in to act on alerts.
- Bold: Wallet‑level delegation copilot — Sector: software (wallets), finance
- Description: Browser extension or wallet plugin that warns of high concentration risk and offers aligned delegate alternatives at the moment of delegation or re‑delegation.
- Tools/Workflows: Client‑side hints; callouts in delegation flows; API for aligned alternatives with rationale.
- Assumptions/Dependencies: Integration partnerships with wallets; consent for off‑chain/on‑chain data use.
- Bold: Community operations insights (working group targeting) — Sector: community management, education
- Description: Use interest clustering to identify communities (e.g., Finance‑Driven, ENS Ecosystem‑Driven) and tailor outreach, onboarding, and explainers to reduce apathy and improve representativeness.
- Tools/Workflows: LLM summarization; cluster dashboards; outreach playbooks.
- Assumptions/Dependencies: Sufficient forum participation; careful handling to avoid stigmatization or echo chambers.
- Bold: Governance risk/compliance reporting for custodians and funds — Sector: finance (compliance), risk
- Description: Incorporate concentration and alignment metrics into internal controls for DAO treasuries, custodians, and governance‑active funds to document due diligence.
- Tools/Workflows: Periodic reports; risk scores; thresholds for re‑delegation or engagement.
- Assumptions/Dependencies: Accepted risk frameworks; legal clarity around DAO governance responsibilities.
- Bold: Attack surface monitoring (rich‑get‑richer spikes/anomaly detection) — Sector: security (web3), analytics
- Description: Detect sudden delegation inflows or coordination toward few hubs; correlate with forum activity and vote timing to flag potential manipulation or bribery.
- Tools/Workflows: Time‑series anomaly detection; event correlation; incident runbooks.
- Assumptions/Dependencies: Baselines per protocol; access to relevant off‑chain signals (e.g., forum posts, social announcements).
- Bold: Open dataset and reproducibility track for researchers — Sector: academia
- Description: Use the released code/data to benchmark delegation mechanisms, alignment metrics, topic taxonomies, and graph analyses across DAOs.
- Tools/Workflows: Versioned datasets; documentation; evaluation scripts; challenge tasks.
- Assumptions/Dependencies: Timely public release; permissive licensing; data ethics.
- Bold: Third‑party analytics products (Dune/Subgraph apps) — Sector: software/analytics, finance
- Description: Publish reusable queries and dashboards tracking delegation networks, power concentration, and alignment indicators for specific protocols.
- Tools/Workflows: Dune dashboards; public APIs; scheduled refreshes.
- Assumptions/Dependencies: Query cost limits; maintenance commitment; protocol‑specific nuances.
Long‑Term Applications
- Bold: Alignment‑aware delegation protocol upgrades — Sector: software (blockchain governance), mechanism design
- Description: Embed alignment metadata and diversification constraints into governance contracts (e.g., multi‑delegate splitting guided by interest alignment and anti‑concentration caps).
- Tools/Workflows: Smart contract changes; audits; simulation/backtesting; formal mechanism design.
- Assumptions/Dependencies: Community consensus for upgrades; economic analysis of incentives; mitigation of sybil/manipulation risks.
- Bold: Privacy‑preserving alignment (ZK‑backed recommenders) — Sector: cryptography/privacy, software
- Description: Use zero‑knowledge proofs to attest to interest categories without revealing raw posts or identities, enabling private, alignment‑based suggestions and routing.
- Tools/Workflows: ZKP circuits for interest attestations; decentralized recommendation infrastructure; DID/VC integrations.
- Assumptions/Dependencies: Efficient proofs; secure taxonomy commitments; UX for consent and revocation.
- Bold: Delegation anti‑concentration policies and standards — Sector: policy/regulation, industry standards
- Description: Develop best‑practice guidelines for fair delegate discovery (e.g., transparency on ranking criteria, caps, diversity prompts) and disclosures for governance platforms.
- Tools/Workflows: Standards documents; compliance checklists; independent audits/certifications.
- Assumptions/Dependencies: Regulator recognition of DAO governance; platform willingness to adopt; cross‑jurisdiction alignment.
- Bold: Cross‑DAO meta‑governance routing — Sector: software (aggregators), finance
- Description: A multi‑protocol layer that helps token holders distribute delegations across DAOs based on their interest profile, minimizing correlated concentration.
- Tools/Workflows: Multi‑chain data ingestion; unified interest profiles; portfolio‑style delegation optimizer.
- Assumptions/Dependencies: API access across platforms; cross‑chain identity resolution; incentives for users to diversify.
- Bold: Reputation systems for delegates (alignment consistency and responsiveness) — Sector: platform governance, analytics
- Description: Composite scores that reward alignment fidelity over time, proposal engagement quality, and responsiveness to delegators—surfaced in discovery UIs.
- Tools/Workflows: Scoring models; anti‑gaming safeguards; periodic recalibration.
- Assumptions/Dependencies: Agreed metrics; guardrails against strategic signaling; fairness audits.
- Bold: Corporate proxy voting alignment for retail investors — Sector: finance (public markets), compliance
- Description: Adapt interest extraction and alignment mapping to suggest proxy advisors or vote recommendations that reflect retail investors’ priorities in shareholder meetings.
- Tools/Workflows: Investor questionnaires; mapping to board agenda categories; broker/platform integrations.
- Assumptions/Dependencies: Access to proxy statement semantics; regulatory acceptance; investor privacy.
- Bold: Civic e‑governance and participatory budgeting — Sector: government/civic tech
- Description: Use interest‑aligned proxy suggestions in municipal or NGO decision processes to improve participation and representation while controlling concentration risks.
- Tools/Workflows: Public consultation portals; alignment questionnaires; transparent explainability for recommendations.
- Assumptions/Dependencies: Identity verification; strong privacy protections; inclusive taxonomies.
- Bold: Token distribution and incentive reform — Sector: protocol economics, governance
- Description: Redesign emission schedules, delegation incentives, or voting weights (e.g., conviction/quadratic voting hybrids) guided by concentration analytics and alignment goals.
- Tools/Workflows: Agent‑based simulations; historical backtests; governance proposals with staged pilots.
- Assumptions/Dependencies: Robust modeling of strategic behavior; careful rollout to avoid shocks; community support.
- Bold: DeFi insurance and lending risk pricing by governance health — Sector: finance (DeFi), insurance
- Description: Incorporate governance concentration/misalignment scores into protocol risk models to price cover and collateral factors more accurately.
- Tools/Workflows: Risk oracles; actuarial models with governance inputs; disclosures to policyholders/borrowers.
- Assumptions/Dependencies: Historical linkage between governance health and incident risk; avoidance of pro‑cyclical penalties.
- Bold: Ethical data linkage standards (consent‑first identity resolution) — Sector: privacy standards, web identity
- Description: Establish norms and tooling for opt‑in linking of forum profiles, ENS names, and wallet addresses using DIDs/VCs, with revocation and selective disclosure.
- Tools/Workflows: Consent workflows; verifiable credentials; privacy policy templates; audits.
- Assumptions/Dependencies: Community trust; interoperability across platforms; safe defaults.
Notes on Cross‑Cutting Assumptions and Dependencies
- Identity resolution quality is pivotal; mislinkages can distort alignment and recommendations.
- LLM‑based topic extraction must be monitored for bias and drift; periodic evaluation against human labels is necessary.
- Privacy, consent, and transparency are foundational for user and platform adoption; explainability should accompany any recommendation.
- Economic and strategic behavior can adapt to new mechanisms; simulations and pilots help avoid unintended concentration or gaming.
- Platform and community buy‑in determines deployment speed; many long‑term items require governance changes or standardization efforts.
Glossary
- ABI (Application Binary Interface): A specification that defines how to encode/decode and invoke smart contract functions on a blockchain. "obtained by parsing contract ABIs via Etherscan"
- Archive node: A blockchain node that stores the full historical state, enabling queries about any past block. "we deployed Ethereum and Arbitrum archive nodes"
- Archival data: Complete historical blockchain data used to reconstruct past actions and states. "We leverage full Ethereum archival data to reconstruct governance actions"
- Clustering: In network analysis, the tendency for nodes to form tightly knit groups (often measured by the clustering coefficient). "Clustering and transitivity are near zero"
- DAO (Decentralized Autonomous Organization): A blockchain-based organization governed by token holders through on-chain rules and votes. "Decentralized Autonomous Organizations (DAOs) aim to enable participatory governance"
- DAPPs (Decentralized Applications): Applications that run on a blockchain without centralized control. "the DAPPs that run atop a blockchain"
- Degree assortativity: A network metric indicating the correlation between the degrees of connected nodes (negative implies high-degree nodes connect to low-degree nodes). "Degree assortativity is negative in all cases"
- Delegation graphs: Directed graphs representing who delegates voting power to whom in governance systems. "delegation graphs are extremely sparse"
- Delegates: Participants who receive delegated voting power and vote on behalf of others. "Delegates are the recipients of delegated voting power."
- Delegators: Token holders who assign their voting power to another participant instead of voting directly. "A delegator is a token holder who chooses not to vote directly"
- Dendrogram: A tree diagram visualizing the hierarchical structure produced by clustering. "the resulting dendrogram is shown in Figure"
- DEX (Decentralized Exchange): A blockchain-based exchange that enables peer-to-peer trading without intermediaries. "Decentralized Exchange (DEX)"
- Disassortative: Describes networks where high-degree nodes tend to connect to low-degree nodes. "a hallmark of disassortative 'hub-and-spoke' organization"
- ENS (Ethereum Name Service): A naming system mapping human-readable names to blockchain addresses. "ENS exhibits the largest scale"
- ENS registry: The smart contract system that records ENS name ownership and related events. "This process relied on ENS registry events"
- Gini coefficient: A measure of inequality (0 equals perfect equality; 1 equals maximal inequality). "Gini coefficients above $0.99$"
- Governance capture: A situation where decision-making is dominated by a small group, undermining representativeness. "including governance capture, protocol ossification, and reduced adaptability"
- GPT (Generative Pre-trained Transformer): A family of LLMs used for text understanding and generation. "GPT-based prompts map these features into interpretable voter interest categories"
- Largest undirected connected component (UG LCC): The biggest subset of nodes that are mutually reachable when ignoring edge directions. "largest undirected connected component (UG LCC)"
- Liquid democracy: A voting system where individuals can either vote directly or delegate their vote to a representative, with delegations being fluid. "something like a liquid or representative democracy"
- LLM: A machine learning model trained on large text corpora to understand and generate natural language. "using LLM techniques to extract topics, sentiment, and ideological signals."
- Nakamoto coefficient: The minimum number of entities required to collude to control a given share (e.g., 33% or 50%) of a system’s resources. "The corresponding Nakamoto coefficients confirm this extreme imbalance"
- Off-chain: Data or processes that occur outside the blockchain (e.g., forums, social platforms). "off-chain discussions from 14 DAO forums"
- On-chain: Data or processes recorded directly on the blockchain. "on-chain data from five major protocols"
- Protocol ossification: The reduced ability of a protocol to adapt or change due to entrenched power structures or rigid governance. "including governance capture, protocol ossification, and reduced adaptability"
- Reciprocity: In directed networks, the extent to which edges are mutual between pairs of nodes. "Reciprocity is negligible ($0.0001$)"
- Rollup: A Layer 2 scaling technique that processes transactions off-chain and posts compressed data on-chain. "Layer 2 Scaling (Rollup)"
- Self-delegation: When a token holder delegates their voting power to their own address. "Self-loops (u→u) represent self-delegation."
- Star-forest topology: A network structure composed of many star-shaped subgraphs centered on hubs. "further supporting a star-forest topology"
- t-SNE: A technique for dimensionality reduction that visualizes high-dimensional data in two or three dimensions. "using t-SNE"
- TimeLock: A governance component that queues and delays execution of approved proposals to allow for review or cancellation. "Emitted when a proposal is executed in the TimeLock."
- Transitivity: A network property related to the likelihood that two neighbors of a node are also connected (global clustering). "Clustering and transitivity are near zero"
- Voting Bloc Entropy (VBE): A metric proposed to quantify the decentralization effects and risks of coordinated voting groups. "propose VBE as a metric to quantify the decentralization effects of voting blocs"
- Ward’s method: A hierarchical clustering criterion that minimizes within-cluster variance at each merge. "We employed Ward’s method as it minimizes within-cluster variance"
- Weakly connected components (WCCs): Subsets of a directed graph where replacing directed edges with undirected edges makes the subgraph connected. "weakly connected components (WCCs)"
- Zero-knowledge proofs (ZKPs): Cryptographic methods allowing one party to prove knowledge of a statement without revealing the statement itself. "based on ZKPs"
Collections
Sign up for free to add this paper to one or more collections.