{"id":"b39f4832-cebc-4467-ac8e-f2b977c878f8","arxiv_id":"2501.13377","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A statistically validated network of shared DeFi governance token holdings reveals persistent, institution-dominated links between protocols.","lead":"This paper maps how a small set of addresses simultaneously hold governance tokens in multiple DeFi protocols, using a network method borrowed from finance. It finds these cross-protocol holders are mostly institutional investors and that their influence dilutes during speculative market booms.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section 3.7 permutation controls are not specified to hold either token; if sampled from all addresses, the significance markers in Figures 3 and 8 are guaranteed by construction, undermining the statistical validation but not the descriptive findings.","rationale":"The reader's weakest_assumption correctly identifies the underspecified permutation control as the most serious methodological flaw. The concern is load-bearing because the paper uses the permuted-control significance markers to single out 'consistently significant' links and 'statistically meaningful' directional influence; if the control group is not matched on token holdings, those markers are guaranteed and carry no information. However, the paper's central descriptive findings - internal influence up to 34%, small link sizes, high Gini coefficients, and institutional majority in Figure 7 - do not depend on the permutation test and remain plausible. The paper also explicitly acknowledges that holdings do not consistently reflect voting power, and the strongest claim is phrased as 'potential to exert' control, so the overall conditional verdict is appropriate. A concrete check of the repository code can settle whether the control group is properly constructed; if it is not, the paper needs a reanalysis of the statistical claims but not a rejection of the descriptive contribution.","tokens_in":22419,"tokens_out":9986,"duration_ms":775890,"concrete_test":"Inspect the public GitHub repository (https://github.com/xm3van/reasearch-project-erc20-governance) for the code sampling A_control in Section 3.7. If controls are drawn uniformly from Ethereum addresses or from all token-holding addresses without requiring them to hold at least one of ti or tj, replicate Figures 3 and 8 with controls sampled from the union of holders of ti and tj (or otherwise matched on token-holding status) and recompute the significance markers. If the markers largely disappear, the statistical validation claims in Section 4.2 are unsupported as written, while the descriptive concentration numbers remain.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3.7 defines the control group A_control only as 'a set of randomly selected addresses, matched in size' to the link-defining set A_i,j. The internal influence metric (Eq. 4) is zero for any address that holds neither ti nor tj. If A_control is drawn uniformly from the Ethereum address space or even from all token-holding addresses without conditioning on holding at least one of ti or tj, then a permutation test that shuffles the combined set will almost always keep the token holdings concentrated in the original A_i,j group, making every difference statistically significant by construction. The paper never states that controls are matched on token holdings, so the significance markers in Figures 3 and 8 cannot be interpreted as evidence that link-defining addresses are unusually influential. This does not invalidate the descriptive values (e.g., up to 34% internal influence, institutional majority in Figure 7), which stand on their own, but it removes the inferential layer used to identify 'consistently significant' links and to claim 'statistically meaningful' directional influence in Figure 8. The central claim about concentration and institutional dominance is primarily descriptive, so the error is not fatal if the permuted-control claims are downgraded or reanalyzed.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies cross-protocol governance concentration in DeFi by constructing a Statistically Validated Network (SVN) of governance tokens based on shared holder addresses across 18 monthly snapshots from January 2021 to June 2022. For each validated token pair, it defines the set of 'link-defining' addresses that hold both tokens and measures their 'internal influence' and 'directional influence' as fractions of token supply, together with link size, Gini coefficients, median wealth, and label-based entity composition. The authors report persistent influential links, internal influence values up to 34% of combined supply, a majority of internal influence held by institutional addresses, and negative correlations between internal influence and TVL changes. The paper also includes a sensitivity analysis and makes code and data publicly available.","tokens_in":22641,"tokens_out":6535,"duration_ms":57020,"significance":"If the descriptive findings hold, this is a useful contribution to the understudied topic of cross-protocol governance control in DeFi: most prior work analyzes token distributions within individual protocols, whereas this paper explicitly examines shared holders across protocols and quantifies the concentration of governance-relevant holdings in small address sets. The paper is transparent about data sources, presents a sensitivity analysis, publishes its code and data, and acknowledges the gap between token holdings and actual voting power. The main weaknesses are in the statistical validation layer: the permutation control group in §3.7 is not specified to hold either token, which undermines the significance markers in Figures 3 and 8, and the TVL correlations in Figure 9 are reported without multiple-comparison correction. These issues affect the inferential claims but not the raw descriptive concentration values, so the paper can be revised rather than rejected.","major_comments":[{"comment":"The control group Acontrol is defined in §3.7 only as 'a set of randomly selected addresses, matched in size' to Ai,j, with no statement that controls hold at least one of the two tokens ti or tj. Because Internal Influence (Eq. 4) and Directional Influence (Eq. 5) are exactly zero for an address holding neither token, a permutation test that shuffles the combined set Ai,j ∪ Acontrol will almost always reject the null for any Ai,j with positive token holdings. As written, the significance markers in Figures 3 and 8 do not provide evidence that link-defining addresses are unusually influential; they largely reflect the fact that random addresses hold none of the relevant tokens. Please specify the sampling universe of Acontrol and re-run the permutation tests with controls matched on token-holding status (e.g., addresses holding at least one of the two tokens), or remove the inferential language and present the values as descriptive quantities.","section":"§3.7, Eqs. (4)–(5), Figs. 3 and 8"},{"comment":"The correlation analysis between internal influence and TVL percentage change reports 18 link-level tests without any multiple-comparison correction. At α=0.05, two of the 18 correlations are significant (CVX-SUSHI and YFI-LDO), which is close to the number of false positives expected by chance, so the statement that 'few correlations achieve statistical significance' is not informative and the negative-correlation result is not robust as presented. Please apply a correction such as Benjamini-Hochberg or explicitly label Figure 9 as exploratory. This point is load-bearing for the Discussion's claim that internal influence shifts with speculative market cycles.","section":"§4.2.1, Fig. 9, Table D.1"}],"minor_comments":[{"comment":"The Internal Influence metric is defined as the arithmetic mean of the fractional holdings in the two tokens; because token supplies differ by orders of magnitude, this average can be dominated by the smaller-supply token. Please state the rationale for the arithmetic mean rather than, for example, a supply-weighted average.","section":"§3.6, Eq. (4)"},{"comment":"The Jaccard similarity in Figure 1 is computed on token-pair links, so the sentence that 'subsets of addresses contributing to the links are likely the same over time' is not directly supported by this figure. Please add an address-level overlap analysis or soften this inference.","section":"§4.1, Fig. 1"},{"comment":"The text says the TVL correlations are 'adjusted for auto-correlation by differencing' but does not describe the adjustment procedure; the reference to 'Appendix Table D.1' is incomplete. Please spell out the differencing/detrending steps and how the reported Durbin-Watson and Ljung-Box statistics were used.","section":"§4.2.1 and Appendix D"},{"comment":"The label-based institutional finding depends on third-party labels from partly closed sources. Please report the number of addresses labelled, the coverage relative to the link-defining sets, and the rate of conflicts during triangulation, so readers can assess label quality.","section":"§3.3 and Fig. 7"},{"comment":"There are several typos and grammatical errors, for example 'the respective addresses already influential within existing Defi Token may were aware' in §4.2, 'governance right a predominantly held by few institutional investor' in §5, and 'favourtism' in §5. A careful proofreading pass is needed.","section":"Throughout"},{"comment":"The phrase 'identify influential addresses that shape these connections' is somewhat circular: the link-defining addresses Ai,j are defined as the addresses holding both tokens, so the Internal Influence metric is computed on the very set that defines the link. The paper should clarify that the concentration is a property of the overlap set rather than an independent causal finding.","section":"Abstract and §3.5"}],"recommendation":"major_revision","confidential_remarks":"The paper is likely salvageable: the descriptive concentration results are valuable and appear technically sound, but the permutation-test control group in §3.7 is underspecified and, as written, invalidates the significance claims in Figures 3 and 8. If the authors cannot re-sample controls with token-holding status, they should explicitly downgrade those claims to descriptive statements. The TVL correlation result also needs multiple-comparison correction or explicit exploratory framing."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The cross-protocol angle is the real news: instead of asking who controls one DAO, they ask which addresses hold tokens in several DAOs at once, and they build a statistically validated network on shared holdings. The descriptive finding is credible: a handful of small address sets account for the token overlap between major DeFi protocols, internal influence reaches roughly 34% of combined supply, and institutional labels dominate most links. That is worth knowing.\n\nWhat the paper does well: data collection and labeling are documented, code and data are on GitHub, sensitivity analysis on the supply threshold is included, and the authors are candid in the limitations paragraph that holdings do not consistently equal voting power. They also connect specific links to known mechanics, like the Convex/Curve relationship, rather than leaving the networks unexplained.\n\nThe soft spot is the permutation test in Section 3.7. The control group is described only as randomly selected addresses matched in size to the link-defining set. If those controls are drawn from the general address space without conditioning on holding at least one of the two tokens, then any shuffle will almost always put token holdings back in the original group, and the significance markers in Figures 3 and 8 are essentially guaranteed by construction. The paper never says controls are matched on token holdings. That weakens the inferential layer as reported. The descriptive numbers do not depend on this test, and the central concentration claim is descriptive, so the paper survives, but the statistical validation language needs to be reworked or the control design needs to be fixed.\n\nA secondary issue: the TVL correlations in Figure 9 involve 18 tests without multiple-comparison correction, and a couple of p<0.01 results would not survive a strict Bonferroni threshold. The autocorrelation adjustments are there, so this is fixable.\n\nThe label analysis is the least precise part because the label sources are partly proprietary, but the authors acknowledge possible mislabeling and the institutional majority is consistent with the median wealth figures. There are no invented entities and no hidden free parameters beyond the usual thresholds, and the sensitivity analysis covers the main one.\n\nBottom line: this is a solid empirical contribution with an underspecified control group. It deserves peer review as a serious paper; the referee should ask for the permutation control to be described precisely and re-run with token-holding controls. I would cite it for the descriptive cross-protocol concentration result.","headline":"A credible descriptive map of cross-protocol governance-token concentration, with an underspecified permutation control that should be fixed before the significance claims are trusted.","tokens_in":23183,"tokens_out":2398,"would_cite":true,"duration_ms":22047,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper shows that across 35 validated links between DeFi governance tokens, a small set of link-defining addresses — mostly institutional investors — holds up to 34% of combined token supply, so a coordinated group could influence…","keywords":["Decentralised Finance","governance tokens","cross-protocol influence","statistically validated networks","token concentration","institutional investors","Ethereum","DAOs"],"falsifier":"Re-run the permutation test with control addresses sampled only from addresses that hold at least one of the two tokens in a link, while keeping the control group the same size as the link-defining set. If the significance markers in Figures 3 and 8 disappear or weaken substantially, then the reported concentration is an artifact of comparing token holders with random Ethereum addresses rather than a property of the holders themselves.","tokens_in":22211,"feed_emoji":"🗳️","tokens_out":7671,"duration_ms":51252,"temperature":0.7,"pith_summary":"The paper asks whether governance power in Decentralised Finance is concentrated not just inside single protocols but across protocols, through addresses that hold governance tokens in several communities at once. It builds a statistically validated network of shared token holdings among 15 DeFi governance tokens and looks at the addresses that define each link. The result is that these link-defining groups are small, hold a high fraction of combined token supply — up to 34% in places — and are mostly labelled as institutional investors. For most links, institutions hold the majority of internal influence, so a coordinated block of addresses could in principle weigh on decisions in more than one protocol at the same time. The authors also find that influence is diluted during speculative upswings, when new buyers enter and core holders' share shrinks.","feed_headline":"Most cross-protocol DeFi governance sits with institutions","feed_subtitle":"Shared token holdings reveal small address groups able to sway votes in several protocols at once.","key_machinery":"The central object is the Statistically Validated Network (SVN) projection of a bipartite graph whose two node sets are governance tokens and the Ethereum addresses holding them. Links between tokens are kept only when the number of addresses holding both tokens is larger than expected under a hypergeometric null model after Bonferroni correction; the addresses that actually hold both tokens of a link are called link-defining addresses $A_{i,j}$. The argument is carried by two measures computed on those addresses: internal influence, the average fraction of token supply held by $A_{i,j}$ across the two tokens, and directional influence, the share of one token's supply held by the link-defining set. These measures translate shared holdings into a concrete statement about how much voting weight a small group could coordinate.","core_discovery":"On the paper's own terms, the central discovery is that cross-protocol governance influence is real and concentrated: statistically validated links between governance tokens are formed by small sets of link-defining addresses, and those addresses hold a disproportionate share of both tokens. Across the 35 validated links studied, internal influence reaches up to 34% of the combined token supply, and the Gini coefficient of influence within most links is above 0.7, meaning a few addresses dominate even inside the linking group. Label analysis shows institutional addresses hold the majority of internal influence on most links, while protocol contracts, vesting contracts, and liquidity pools dominate particular pairs such as CVX-CRV. Links persist over time, and directional influence is usually concentrated on one token in the pair, giving the linking addresses asymmetric control across the two communities. The paper reads these patterns as evidence that token-based governance contains an underexplored risk vector: the same holders can influence several protocols at once.","pith_inferences":["The authors measure holdings, not votes; linking the holder network to on-chain proposal votes would show whether the potential coordinated control is ever exercised.","If the control group for the permutation test is not required to hold at least one of the tokens, re-running the comparison with token-holding controls would show how much of the reported significance is an artifact of sampling ordinary Ethereum addresses.","The same network construction could be run on other chains or on multi-chain governance tokens, where shared holders may be even harder to attribute to a single community.","Joining holdings data with vesting schedules and contract labels could separate structural concentration (tokens locked in contracts) from discretionary concentration (wallets that can actively vote), which have different implications for governance risk."],"forward_implications":["If the link-defining addresses acted together, they could simultaneously shape governance decisions in both token communities of a link, because their combined holdings pass substantial voting thresholds in several cases.","Single-protocol studies of governance concentration underestimate the problem: the same institutions appear on both sides of validated links, so cross-protocol concentration is a distinct layer of control.","The dominance of institutional labels implies that profit-oriented entities, not retail users, hold most of the cross-protocol governance weight, creating potential conflicts with community interests.","Because directional influence is usually concentrated on one token per link, control between paired protocols is asymmetric — the CVX-CRV case shows one protocol dependency driving another's governance weight.","During market upswings, rising TVL dilutes the internal influence of core holders, meaning speculative inflows temporarily shift governance power away from long-term community members."],"supporting_citations":[{"why":"Supplies the hypergeometric null model and the statistically validated network construction used to test which token pairs share more addresses than expected by chance.","marker":"[7]"},{"why":"Introduces the application of statistically validated networks to portfolio overlaps and motivates the focus on address sets that define validated links.","marker":"[8]"},{"why":"Provides the token selection criteria and the within-protocol concentration baseline that this paper extends to the cross-protocol setting.","marker":"[3]"},{"why":"Supplies the prior finding that a few addresses control most DAO voting power, which this paper generalizes across protocol pairs.","marker":"[5]"},{"why":"Offers the closest prior evidence on governance influence across DAOs, which this paper explicitly builds on to study cross-protocol control.","marker":"[57]"},{"why":"Documents the Ethereum ETL pipeline used to retrieve monthly governance token balance tables from an Erigon node.","marker":"[6]"},{"why":"Explains the Convex-Curve dependency that the paper uses to interpret the high directional influence of CVX on CRV.","marker":"[36]"}],"fun_headline_variants":["Cross-protocol DeFi votes hinge on tiny address groups","Few addresses dominate governance across DeFi protocols","Institutional wallets steer multiple DeFi governance votes","Shared tokens reveal concentrated DeFi governance power","DeFi governance concentration spans protocol boundaries"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The paper's significance tests assume that the randomly chosen comparison addresses are comparable to the link-defining addresses; because the paper never states that the comparison addresses hold at least one of the two tokens, the reported gaps could be nearly automatic if the comparison group came from ordinary Ethereum addresses.","fun_headline_variants_meta":{"raw":{"variants":["Cross-protocol DeFi votes hinge on tiny address groups","Few addresses dominate governance across DeFi protocols","Institutional wallets steer multiple DeFi governance votes","Shared tokens reveal concentrated DeFi governance power","DeFi governance concentration spans protocol boundaries"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000753,"raw_usage":{"total_tokens":3313,"prompt_tokens":869,"completion_tokens":2444,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":485,"completion_tokens_details":{"reasoning_tokens":2375}},"tokens_in":485,"tokens_out":2444,"duration_ms":28221,"temperature":1.0,"reasoning_tokens":2375,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T16:00:21.953530+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the permutation test with control addresses sampled only from addresses that hold at least one of the two tokens in a link, while keeping the control group the same size as the link-defining set. If the significance markers in Figures 3 and 8 disappear or weaken substantially, then the reported concentration is an artifact of comparing token holders with random Ethereum addresses rather than a property of the holders themselves.","supporting_citations":[{"cited_title":"& Mantegna, R","cited_arxiv_id":null,"evidence_quote":"Supplies the hypergeometric null model and the statistically validated network construction used to test which token pairs share more addresses than expected by chance."},{"cited_title":"Decentralized Finance, Centralized Ownership? An Iterative Mapping Process to Measure Protocol Token Distribution","cited_arxiv_id":"2012.09306","evidence_quote":"Provides the token selection criteria and the within-protocol concentration baseline that this paper extends to the cross-protocol setting."},{"cited_title":"The Governance of Decentralized Autonomous Organizations: A Study of Contributors’ Influence, Networks, and Shifts in Voting Power","cited_arxiv_id":null,"evidence_quote":"Offers the closest prior evidence on governance influence across DAOs, which this paper explicitly builds on to study cross-protocol control."},{"cited_title":"& D5 team Ethereum ETL","cited_arxiv_id":null,"evidence_quote":"Documents the Ethereum ETL pipeline used to retrieve monthly governance token balance tables from an Erigon node."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Explains the Convex-Curve dependency that the paper uses to interpret the high directional influence of CVX on CRV."}],"review_version":1}