{"id":"201ea5cf-0901-4f06-857d-963036d8ec69","arxiv_id":"2504.12859","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":2,"one_line_summary":"The authors prove that square-root and power-law stake weighting reduce Gini-measured voting inequality, though the claimed strict improvement in the Nakamoto coefficient is false in general.","lead":"This paper analyzes quadratic voting schemes for blockchain governance and proves that square-root weighting lowers the Gini coefficient of voting power. It also proposes a generalized power-law voting rule with an algorithm to cap the largest stakeholder's influence.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Nakamoto coefficient strict increase (Prop. 5) fails because top-k share dominance gives only non-strict inequality; equating thresholds can make the number of needed voters equal.","rationale":"The reader identified the same load-bearing weakness: the step from inequality (11) to the strict increase of the Nakamoto coefficient in Proposition 5 does not hold. My independent check of the proof confirms that Lemma 2 (summation by parts) yields only A_k <= B_k for the relevant cumulative sums, and equality can occur in the top-k partial sums even for k < n whenever the largest k elements have identical relative weights; the non-strict bound cannot be upgraded to strict at the specific threshold k*. The two-voter counterexample is decisive: for stakes (1,100), the largest voter controls more than 51% under both linear (100/101) and quadratic (10/11) schemes, so N_LV(a) = N_QV(a) = 1 for every a in [0.51, 100/101]. This directly falsifies the abstract and conclusion claims about the Nakamoto coefficient. The paper contains other substantive issues (the utility maximization algebra in Section 4 has sign and factor errors, the QV-3 Sybil claim is contradicted by the unsplit-stake design, and the abstract's ordering of Type 1 and Type 2 QV conflicts with Definition 1), but Proposition 5 is the single most load-bearing failure because it is the formal core of the advertised decentralization result. Since the headline claim is false, REJECT is appropriate; a repair would require restating the result as non-strict (N_gammaV(a) >= N_LV(a)) and replacing the universal strict-improvement claim with bounds that hold only outside the equal-coefficient cases.","tokens_in":21739,"tokens_out":2053,"duration_ms":18904,"concrete_test":"Compute both sides of inequality (11) for the two-voter stake distribution (1,100) with gamma = 1/2 and threshold a = 0.51. The top-1 linear normalized share is 100/101 and the top-1 quadratic share is 10/11; both exceed a, so N_LV(0.51) = 1 and N_QV(0.51) = 1, contradicting Proposition 5's strict inequality. A broader brute-force check over all stake vectors with n=3 and thresholds a in {0.51, 0.67, 0.90} would show how frequently the strict increase fails.","verdict_should_be":"REJECT","load_bearing_attack":"The headline claim that QV (Types 2 and 3) and GPV strictly increase decentralization as measured by the Nakamoto coefficient rests on Proposition 5 (Section 3.2), which states N_{\\gamma V}(a) > N_{LV}(a) for every threshold a. The proof argues from inequality (11), claiming that for every k the normalized gamma top-k share is strictly below the top-k share for linear voting. However, inequality (11) is actually a non-strict dominance statement: the sum of the top-k normalized shares under gamma-power weighting is less than or equal to the corresponding sum under linear weighting, with equality possible whenever k includes all n voters or when the two weight vectors coincide. The intended conclusion N_{\\gamma V}(a) > N_{LV}(a) requires a strict inequality at the exact threshold index where the linear top-k sum first reaches a; non-strict dominance permits the gamma share to already be at or above a at that same k, giving equal Nakamoto coefficients. The two-voter example with stakes {1,100} and a = 0.51 shows the failure: the linear top-1 share is 100/101 > 0.51, while the quadratic top-1 share is 10/11 > 0.51, so both Nakamoto coefficients equal 1. Thus Proposition 5 is false as stated, and the central claim that QV and its generalizations strictly increase the Nakamoto coefficient is unsupported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies Quadratic Voting (QV) in blockchain governance, distinguishing three QV types depending on whether the stake is split and whether the square root is applied before or after splitting, and introduces a generalized 'gamma-power voting' (GPV) scheme. The authors prove properties of the relative voting ratio (RVR) under QV-2 and QV-3, show that the Gini coefficient decreases for QV-2 and GPV compared to linear voting, and claim that the Nakamoto coefficient strictly increases for every threshold. They also analyze utility maximization for voters who vote after observing others, concluding that privacy-preserving voting protocols are necessary, and propose an algorithm to calibrate the gamma parameter to limit the influence of the largest stakeholders.","tokens_in":21983,"tokens_out":9256,"duration_ms":94044,"significance":"If the main claims were correct, the paper would provide a useful formal toolkit for comparing voting schemes by decentralization metrics and for tuning a family of concave weighting schemes. The RVR results (Propositions 1-2) and the Gini inequality (Propositions 3-4) are elementary but correct in substance, with the gamma-power parametrization being a natural extension. However, the headline result that QV and GPV strictly increase the Nakamoto coefficient (Proposition 5) is false, and the utility-maximization section contains algebraic and logical errors. Because these issues affect the central contribution, the paper in its current form is not publishable, even though parts of the formal framework could be salvaged after substantial revision.","major_comments":[{"comment":"The claimed strict inequality N_{γV}(a) > N_{LV}(a) for every threshold a is false. The proof through Lemma 2 with the indicator weights y_i yields only the non-strict dominance ∑_{i=n-k+1}^n (s(i))^γ / ∑_{i=1}^n (s(i))^γ ≤ ∑_{i=n-k+1}^n s(i) / ∑_{i=1}^n s(i), with equality at k=n. Even if this dominance were strict, it would not imply strict increase of the Nakamoto coefficient: when the linear top-k share first reaches a, the gamma top-k share may already be at or above a. For stakes {1,100} and a=0.51, both the linear top-1 share (100/101) and the quadratic top-1 share (10/11) exceed 0.51, so both Nakamoto coefficients equal 1. The counterexample directly disproves Proposition 5 as stated, and with it the abstract claim that Types 2 and 3 QV and their generalizations strictly enhance decentralization as measured by the Nakamoto coefficient.","section":"Section 3.2, Proposition 5 and inequality (11)"},{"comment":"The Lagrange multiplier derivation for QV-2 is incorrect. For the constraint √s(i) = ∑_r s_r^{(i)}, the derivative of the Lagrange function (17) with respect to s_r^{(i)} is -λ, not -2λ, so the stationarity condition should be π_r^{(i)}(b_r-a_r)/(s_r^{(i)}+b_r)^2 = λ. Consequently the expression for λ and equation (16) do not follow from the stated Lagrangian. Similarly, the second derivative ∂²L/∂s_r^{(i)}∂λ equals -2 in the paper, but for this linear constraint it should be -1. These errors undermine the utility-maximization and last-voter manipulation arguments in Section 4.2.","section":"Section 4.2, Proposition 6 and equations (16)-(17)"},{"comment":"The utility-maximization proof for QV-1 is also incomplete: equation (14) is not the stationarity condition derived from the Lagrangian (15) except under a restrictive proportionality assumption on the A_r = π_r^{(i)}(b_r-a_r)/(s_r^{(i)}+b_r)^2 terms, and the paper does not justify that assumption. Moreover, Lemma 3 asserts that a symmetric matrix with negative diagonal entries, negative off-diagonal entries in the last row/column, and a zero bottom-right entry is negatively definite; this is false in general, as such a matrix need not be negative definite even for n=2. The negative-definiteness of the Hessian is therefore not established by the given argument.","section":"Section 4.1, Lemma 3 and equation (14)"},{"comment":"The statement 'both types of QV, type 1 and type 2, are vulnerable to Sybil attacks, but QV-3 is not' is false. In QV-3, an attacker with total stake S split into k accounts can contribute k·√(S/k) = √(kS) votes to a proposal, which is strictly larger than √S for k>1. Thus QV-3 is also Sybil-vulnerable; splitting improves an attacker's influence. This claim should be corrected or removed. The decentralization theorems in Section 3 do not rely on it, but it appears as a stated advantage of QV-3 and is currently misleading.","section":"Section 2, paragraph before Proposition 1"},{"comment":"The proof of the Gini inequality contains a normalization error. The text defines b_k = (s(k))^γ / (∑ s(i))^γ and then claims A_n = B_n = 1, but ∑ b_i = ∑ (s(i))^γ / (∑ s(i))^γ, which is not 1 in general. Lemma 2 cannot be applied as written without normalizing b_i by ∑ (s(i))^γ. The Gini statement itself is true and can be proved by the same majorization argument once this normalization is fixed, but the formal proof in the manuscript is not correct as it stands.","section":"Section 3.1, proof of Proposition 3"}],"minor_comments":[{"comment":"There are two different results numbered Proposition 5: the Nakamoto coefficient statement in Section 3.2 and the utility-maximization statement in Section 4.1. The numbering should be corrected throughout.","section":"Throughout"},{"comment":"The text refers to 'QV-5' in the informal discussion following Proposition 2; this should be 'QV-3'.","section":"Section 2.2"},{"comment":"The threshold is written as 'a ∈ 0,1', which should be 'a ∈ (0,1)'.","section":"Section 3.2, Definition 4"},{"comment":"The sum in the utility function is indexed by i over the number of proposals, but the voter index is also i; this creates a notational clash. The proposal index should be r throughout.","section":"Section 4.1, equation (13)"},{"comment":"The phrase 'the maximum of a continuous function is a closed area' should say that the maximum is attained on a compact feasible set; the feasible set is compact when the s_r are nonnegative or when the sphere constraint is used.","section":"Section 4.2"}],"recommendation":"reject","confidential_remarks":"The paper contains an interesting and correct Gini-coefficient proof and a useful generalization to gamma-power voting, but the central Nakamoto-coefficient claim is false, and the utility-maximization section has multiple algebraic errors, including a false lemma. These are not merely presentation issues; they affect the main advertised results. The duplicate proposition numbering and the incorrect Sybil claim suggest that the manuscript needs a thorough reworking before it can be considered for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things. First, the paper is not empty: the RVR propositions, the Gini coefficient proposition for gamma-power weighting, and the algorithm for tuning maximal stake impact to a desired bound are correct and moderately useful. Second, the central claim in the abstract—that QV types 2 and 3 strictly improve the Nakamoto coefficient—is false. The stress-test note is right, and the counterexample is embarrassingly simple: stakes {1,100}, threshold a = 0.51. Linear voting needs 1 voter (100/101 > 0.51); QV also needs 1 voter (10/11 > 0.51). So N_QV = N_LV = 1, not N_QV > N_LV. Proposition 5's proof uses inequality (11), but Lemma 2 only gives non-strict dominance at the threshold index. The paper quietly turns '≥' into '>'. That sinks the abstract's advertised result. This is a load-bearing flaw, not a typo. The utility-maximization section also divides by s(i) instead of sqrt(s(i)) in equation (14) for QV-1, and the claim that QV-3 is not Sybil-vulnerable is backwards—with unsplit stakes, splitting one large stake into many small ones actually increases total voting power because sqrt is concave. On top of that, the abstract contradicts Definition 1 on which QV type applies sqrt before versus after splitting. These are not subtle issues. What the paper does well: Proposition 1 and 2 give clean monotonicity properties of relative voting ratios, and the Gini result (Proposition 3/4) is correctly proven for arbitrary gamma in (0,1). The algorithm in Section 6, choosing gamma to cap the largest stakeholder's transformed impact, is a genuine and practical idea that survives the other mistakes. The related work section is adequate, though it leans heavily on blog posts and informal sources. Who is this for? Governance designers working on DAO or Cardano Catalyst might read Section 6 with profit, but they should ignore the Nakamoto claims until they are reworked. As is, the paper needs major revision. I would not take it, but I also would not desk-reject without a referee: the gamma-power parameterization is worth a serious look, and a competent referee can separate the salvageable core from the false headline. My honest recommendation: reject, but invite a resubmission that drops or replaces Proposition 5, fixes the Lagrange error, and retracts the Sybil claim for QV-3.","headline":"A mix of correct elementary results and a false headline claim: the strict Nakamoto improvement (Prop. 5) fails on a two-voter example, and the paper also gets the Sybil direction and a Lagrange multiplier step wrong.","tokens_in":22541,"tokens_out":3468,"would_cite":false,"duration_ms":34694,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91B12","91B14"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that square-root weighting of blockchain stakes—Types 2 and 3 quadratic voting and any power-law generalization—strictly improves decentralization measured by the Gini and Nakamoto coefficients, while preserving the…","keywords":["quadratic voting","blockchain governance","decentralization","Gini coefficient","Nakamoto coefficient","relative voting ratio","gamma-power voting","stake weighting"],"falsifier":"For stakes $\\{1,100\\}$ and threshold $a=0.51$, the largest linear share is $100/101\\approx0.9901$ and the largest square-root share is $10/11\\approx0.9091$; both exceed $0.51$, so the Nakamoto coefficient is $1$ under both linear and square-root weighting, contradicting the strict inequality $N_{\\gamma V}(a)>N_{LV}(a)$ of Proposition 5.","tokens_in":21546,"feed_emoji":"🗳️","tokens_out":12962,"duration_ms":117866,"temperature":0.7,"pith_summary":"This paper argues that replacing one-coin-one-vote with square-root weighting in blockchain governance makes decision power measurably less concentrated. It distinguishes three flavours of quadratic voting and proves that, in the split-stake Type 2 and the unsplit-stake Type 3, small stakeholders gain relative influence while the largest stakeholders lose it. The same formalism is extended to any power-law stake weighting with exponent between 0 and 1, for which the paper proves lower Gini coefficients, higher Nakamoto coefficients, and the existence of a threshold stakeholder below whom everyone gains influence. Governance designers can tune the exponent to hit a desired cap on the largest participant's impact, such as 'the richest 40% hold no more than 60% of voting credit.'","feed_headline":"Quadratic voting provably raises blockchain decentralization","feed_subtitle":"Square-root weighting shrinks the biggest voter's share and lifts Gini and Nakamoto measures via a tunable exponent.","key_machinery":"The central object is the relative voting ratio $r_g^{(i)}=\\frac{g(s^{(i)})}{\\sum_j g(s^{(j)})}$, and the key mechanism is the power-law stake transformation $T_\\gamma(s)=s^\\gamma$, called gamma-power voting (GPV), with QV as the $\\gamma=1/2$ case. The proof machinery is Lemma 2, a summation-by-parts dominance lemma: if the partial sums of normalized linear shares are at most the partial sums of gamma-normalized shares, then for any nondecreasing sequence of weights the linear weighted sum is at least the gamma weighted sum. Applied to weights $y_i=i$, this yields the Gini decrease; applied to a step function selecting the top $k$ shares, it yields the top-$k$ dominance inequality behind the Nakamoto claim. The threshold-stakeholder behaviour is carried by $\\eta^{(i)}=r_{QV}^{(i)}/r_{LV}^{(i)}$, which is shown to be strictly decreasing in stake.","core_discovery":"The paper's central claim is that Types 2 and 3 quadratic voting, and more generally any gamma-power voting with $0<\\gamma<1$, enhance decentralization as measured by the Gini and Nakamoto coefficients. In the paper's terminology, Type 2 QV splits the stake first and then takes square roots of the parts; Type 3 QV gives the square root of the total stake to each supported proposal. For both, the relative voting ratio of the smallest stakeholder rises and that of the largest falls, while the ordering of influence by stake size is preserved. The paper identifies a threshold stakeholder: those below gain relative voting ratio, those above lose it, and this threshold can be moved by changing the exponent. It further proves a strict decrease of the Gini coefficient, claims a strict increase of the Nakamoto coefficient at every control threshold, and shows that without privacy-preserving cryptographic ballots a late voter can maximize utility against observed results, while Type 1 allows collusion by large stakeholders splitting stakes across proposals.","pith_inferences":["A concrete before/after test in a real DAO switching from linear to quadratic voting should show voters below the threshold gaining relative voting share and those above losing it, with the ordering of influence unchanged.","The same gamma-tuning algorithm could be used as a constitutional guardrail: specify the number $k$ of largest stakeholders and the maximum total influence $\\alpha$ they may control, then solve for $\\gamma$ to enforce that cap.","The dominance proof is not tied to power laws, so the decentralization conclusion should also hold for other sublinear weighting functions such as a capped logarithm; testing one would show whether the power-law form is essential.","Type 3 QV makes voting power depend on how many proposals a voter supports, so participation breadth itself changes influence; that creates both an engagement incentive and a bribery surface the paper does not fully explore."],"forward_implications":["The largest stakeholder's relative voting ratio strictly decreases and the smallest's strictly increases under Type 2 and Type 3 QV, while the order of influence by stake is preserved.","The Gini coefficient of voting-credit distribution is strictly smaller under any gamma-power weighting with $0<\\gamma<1$ than under linear one-coin-one-vote weighting.","The Nakamoto coefficient is strictly larger under gamma-power weighting at every control threshold, so reaching any controlling share requires more stakeholders.","For any desired cap $\\alpha$ between $1/n$ and the largest linear share, there is a unique exponent $\\gamma$ that makes the largest transformed stake share exactly $\\alpha$, enabling algorithmic decentralization tuning.","Deploying QV without privacy-preserving ballots lets voters who vote last optimize their utility against observed results, and Type 1 QV is vulnerable to collusion attacks where large stakeholders split stakes evenly across proposals."],"supporting_citations":[{"why":"Introduces quadratic voting and the utility-maximization framework that Section 4 extends.","marker":"Posner and Weyl (2014)"},{"why":"Defines AQV, the split-stake square-root scheme the paper calls Type 2, and supplies the game-theoretic setup generalized in Section 4.","marker":"Dimitri (2022)"},{"why":"Defines Type 1 QV, the split-stake scheme the paper analyzes for collusion attacks.","marker":"Robey (2022)"},{"why":"Defines Type 3 QV and flags Sybil-attack susceptibility that the paper revisits.","marker":"Bobby (2022)"},{"why":"Establishes the use of Gini and Nakamoto coefficients to measure decentralization in DAO governance.","marker":"Fritsch et al. (2022)"},{"why":"Formalizes the Nakamoto coefficient threshold concept used in Definition 4.","marker":"Srinivasan and Lee (2017)"},{"why":"Provides empirical context on wealth decentralization in blockchain networks that motivates the decentralization metrics.","marker":"Juodis et al. (2024)"}],"fun_headline_variants":["Gamma-root voting reins in whales in blockchain governance","Tunable exponent lets DAOs dial in decentralization","Small stakeholders gain influence under quadratic voting","Quadratic voting shrinks Gini and lifts Nakamoto coefficient","Proof: square-root voting reduces inequality in blockchain ballots"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof that gamma-power weighting strictly increases the Nakamoto coefficient for every threshold assumes that every normalized top-$k$ share under gamma weighting is strictly smaller than the corresponding linear top-$k$ share; if that pointwise dominance fails near the threshold boundary, the strict increase can fail.","fun_headline_variants_meta":{"raw":{"variants":["Gamma-root voting reins in whales in blockchain governance","Tunable exponent lets DAOs dial in decentralization","Small stakeholders gain influence under quadratic voting","Quadratic voting shrinks Gini and lifts Nakamoto coefficient","Proof: square-root voting reduces inequality in blockchain ballots"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000778,"raw_usage":{"total_tokens":3487,"prompt_tokens":1043,"completion_tokens":2444,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":659,"completion_tokens_details":{"reasoning_tokens":2370}},"tokens_in":659,"tokens_out":2444,"duration_ms":17297,"temperature":1.0,"reasoning_tokens":2370,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T12:21:28.943154+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For stakes $\\{1,100\\}$ and threshold $a=0.51$, the largest linear share is $100/101\\approx0.9901$ and the largest square-root share is $10/11\\approx0.9091$; both exceed $0.51$, so the Nakamoto coefficient is $1$ under both linear and square-root weighting, contradicting the strict inequality $N_{\\gamma V}(a)>N_{LV}(a)$ of Proposition 5.","supporting_citations":[],"review_version":1}