{"id":"28840492-e18f-4f67-b853-9829b5210573","arxiv_id":"2505.04136","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Partial abstention can in theory replicate optimal weighted voting with minimal coordination, and explicit vote delegation has inherent epistemic weaknesses.","lead":"This paper uses mathematical voting theory to compare ways of governing DAOs, and argues that letting each token holder abstain on a portion of their votes is often better than letting them transfer votes to delegates. A generalist reader would consult it to see whether wisdom of crowds arguments justify decentralized voting over expert control.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The optimality of partial abstention rests on Assumption 3 (accurate self-assessed competencies); the paper acknowledges the assumption but provides no sensitivity analysis, so the practical strength of the claim is unquantified.","rationale":"The reader's weakest_assumption matches the central vulnerability. Proposition 3.2 is mathematically correct under its ideal assumptions: given exact p_i and common R, abstention fractions induce weights proportional to log-odds, and the proof is sound. However, the claim that partial abstention is a practically strong governance method depends on the accuracy of self-assessed competencies, and the paper provides no quantitative robustness check. The theorem's validity under ideal assumptions is not in question; the gap is between the idealized model and the claimed governance recommendation. The reader's CONDITIONAL verdict is reasonable, though it rests on minor technical issues (incomplete Prop 5.1, typo in Cor 4.4) rather than on this deeper assumption. My stress-test does not change the verdict, but it highlights that the condition for acceptance should include either a robustness analysis or a more modest practical claim. The paper itself is honest about the limitation, which is why this is a conditional concern rather than a fatal flaw.","tokens_in":40075,"tokens_out":15553,"duration_ms":148987,"concrete_test":"Run a Monte Carlo study: draw N=100 voters with true competencies p_i from a logit-normal distribution (e.g., logit(p_i) ~ N(0,1)), and draw reported competencies \\hat{p}_i = logit^{-1}(logit(p_i)+\\epsilon_i) with \\epsilon_i ~ N(0,\\sigma^2) for \\sigma \\in {0.1, 0.3, 0.5}. Compute the probability of a correct decision under the partial-abstention mechanism using \\hat{p}_i (voters exercise w(\\hat{p}_i)/R votes) and compare with (a) the optimal weighted rule using true p_i and (b) unweighted majority voting. If for \\sigma=0.3 the partial-abstention probability falls below the majority-vote baseline, the concern lands. Repeat with N=1000 to check the asymptotic behavior.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Proposition 3.2 shows that if each voter knows their own p_i exactly and all voters know R (or coordinate on a sufficiently large \\tilde{R}), partial abstention reproduces the optimal weighted vote. The load-bearing step is the 'if': the entire construction maps reported competencies into voting weights via t*_i = w_i/R. If a voter's self-assessed p_i is biased, the exercised weight is wrong by the derivative of the log-odds; low-competence voters who overestimate p_i will cast too many votes and can flood the decision, while high-competence voters who underestimate will cast too few and waste signal. The paper's conclusion concedes that 'voters may not have very sound knowledge of their own competencies,' but no result quantifies how sensitive the collective success probability is to such miscalibration. Since the headline claim is that partial abstention is a 'strong governance method,' this unquantified sensitivity is the main vulnerability of the central argument. The theorem itself is valid under the assumptions; the concern is that the assumptions are doing all the work and the paper does not show the mechanism degrades gracefully.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces two epistemic tests for decentralized governance mechanisms, one for independent voter competencies and one for dependent competencies modeled through a canonical list of independent signals. Using the Nitzan–Paroush optimal weighting theorem as an external benchmark, it analyzes partial abstention, direct participation, and transfer delegation. The central positive result is Proposition 3.2, which shows that if each voter knows their own competence and all voters coordinate on a single number R, partial abstention can replicate the optimal weighted vote. Section 4 identifies conditions under which additional direct participation is weakly beneficial and characterizes a 'flooding danger' from low-competence voters. Section 5 argues that transfer delegation, including liquid democracy, faces severe coordination and information requirements, with Proposition 5.3 giving a three-step delegation scheme that achieves optimality in connected star-cluster graphs. The paper concludes by discussing supplementary mechanisms such as futarchy, AI agents, and contestable control.","tokens_in":40284,"tokens_out":11530,"duration_ms":122251,"significance":"If the results are accepted, Proposition 3.2 is an elegant and nontrivial observation: under strong but clearly stated assumptions, partial abstention achieves the same collective outcome as a social planner who directly applies optimal weights, with much weaker coordination than transfer delegation. The paper is commendably transparent about its assumptions and about the limits of its framework, and several proofs (Propositions 3.2, 3.8, 4.1, 5.3) are cleanly written and check out. The external benchmark of Nitzan and Paroush is used correctly, and the paper does not circularly fit parameters to data. The main limitation is that the headline practical claim that partial abstention is a 'strong governance method' rests entirely on Assumption 3, and the paper provides no sensitivity analysis for miscalibrated self-assessed competencies. This makes the theoretical results valuable but the practical inference more fragile than the abstract suggests.","major_comments":[{"comment":"The optimality of partial abstention is entirely conditional on each voter knowing their own p_i exactly, and the paper provides no analysis of what happens when voters misestimate p_i. The conclusion (§7) concedes that 'voters may not have very sound knowledge of their own competencies,' but without a sensitivity bound or numerical exploration, the abstract's claim that 'partial abstention is a strong governance method' is not supported as a practical statement. Please add a formal sensitivity result (e.g., a Lipschitz bound on the collective success probability as a function of individual log-odds errors) or numerical simulations showing how the probability of a correct collective decision degrades with miscalibration, and adjust the wording of the practical claim accordingly.","section":"§3, Proposition 3.2 and Assumption 3"},{"comment":"Proposition 5.1 is explicitly incomplete. The proof only analyzes the coalition pair with the minimum voting-rights gap and then states that 'it is likely that a maximizing algorithm is possible at least by brute force, but we leave that possibility to future work.' As a result, the proposition's statement that conditions (C) and (D) are sufficient for a voter to make transfer delegation decisions that weakly increase the probability of a correct collective decision is not established. Since this proposition is used to argue that optimal transfer delegation requires weight omniscience and creates severe coordination problems, the argument is currently stronger than the proof supports. Please either prove the general maximizing algorithm, restrict the proposition to the pairwise decision that is actually proven, or explicitly label the general claim as a conjecture.","section":"§5.2, Proposition 5.1"},{"comment":"The statement of Corollary 4.4(iii) contains a mathematical typo: the range for which adding two voters decreases the probability of a correct collective decision should be defined by w_i + w_j < ln(phi), not w_i + w_j < phi. The proof and the use of the result in part (iv) rely on ln(phi), and as written the inequality is false in general because phi > ln(phi) for phi > 1. This should be corrected.","section":"§4, Corollary 4.4(iii)"}],"minor_comments":[{"comment":"The explanation that a voter with w_m > tilde R receives an 'effective weight of 1000' could be misunderstood, since each voter has only one voting right after reallocation; the intended normalization (multiplying exercised rights by tilde R) should be stated explicitly to avoid confusion.","section":"§3, Example 3.7"},{"comment":"The proposition is presented without a proof; a short proof or a direct citation to the Nitzan and Paroush theorem applied to signals would improve completeness.","section":"§2.2.3, Proposition 2.2"},{"comment":"The sortition discussion is informal and does not state results as propositions; adding a precise comparative statement (e.g., a bound on the loss from excluding positive-weight voters) would make the section more useful to readers who want to compare mechanisms formally.","section":"§5.4, Sortition"},{"comment":"The reference to the House Financial Services Committee appears twice with the same author name but different years; please format as 2023 and 2024 entries consistently.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is unusually long and combines rigorous formal results with broad speculative sections on AI agents, prediction markets, and contestable control. The core theoretical contribution is sound, but the practical claim about partial abstention needs either tempering or a sensitivity analysis. The editor may also wish to consider whether the informal sections in Section 6 fit within the scope of the journal or should be trimmed to keep the focus on the epistemic framework."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things you should know. First, the paper makes a genuine theoretical point: under the Condorcet/Nitzan-Paroush model, partial abstention—voters exercising only a fraction of their voting rights—can reproduce the optimal weighted voting outcome with far less coordination than transfer delegation. Proposition 3.2 is clean and correct. Second, the practical force of that result is entirely hostage to Assumption 3, that voters can accurately assess their own competence. The author admits this in the conclusion, but the paper never quantifies how badly miscalibration degrades the mechanism. That is the main soft spot.\n\nWhat is actually new: the partial-abstention mechanism (Prop 3.2), the dependence-case algorithm (Prop 3.8), and the star-cluster multi-step delegation result (Prop 5.3). The comparison of participation regimes and the flooding/whale dangers are well drawn, and the paper engages seriously with the liquid democracy literature. The proofs I checked are sound, modulo a typo in Corollary 4.4(iii): the condition should be w_i + w_j < ln(phi), not w_i + w_j < phi. Proposition 5.1 is explicitly incomplete—the maximizing algorithm is left to future work—so it reads as a partial result, but the author flags it clearly.\n\nThe soft spots are real but not disqualifying. Assumption 3 is doing a lot of load-bearing work, and the paper offers no sensitivity analysis; a short section on miscalibration would materially improve it. Assumption 7 (signal numeracy) is even stronger and quietly restricts the dependence-case result. The tone also slightly oversells partial abstention as 'a strong governance method' given the author's own caveats; it is a strong theoretical benchmark under very favorable assumptions.\n\nWho this is for: anyone working on DAO governance, jury theorems, or weighted voting. It deserves a serious referee. My recommendation: send it out, but ask for the typo fix, an explicit sensitivity discussion (even informal), and a clearer separation between the theorem and the practical claim.","headline":"Partial abstention result is real and mostly proven, but the paper's practical claims outrun its strong assumptions—worth a serious referee.","tokens_in":40821,"tokens_out":1696,"would_cite":true,"duration_ms":17148,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Partial abstention beats transfer delegation in DAO votes","keywords":["epistemic voting theory","partial abstention","optimal voting weights","liquid democracy","decentralized governance","Condorcet jury theorem","voter competence","DAO governance"],"falsifier":"Run a binary-choice task with known correct answers, elicit each participant's self-assessed p_i, apply the abstention rule w_i/t*_i = R, and compare collective accuracy against simple majority and the social-planner optimum. The claim is falsified if the abstention rule underperforms a naive rule whenever self-assessed p_i are miscalibrated.","tokens_in":39856,"feed_emoji":"🗳️","tokens_out":6491,"duration_ms":62822,"temperature":0.7,"pith_summary":"This paper asks how decentralized organizations such as DAOs can make decisions that are both decentralized and likely to be correct, and it evaluates governance methods by the probability that they pick a correct answer when one exists. Its central result is that partial abstention, voters exercising only a portion of their voting rights, can reproduce the optimal weighted vote almost without coordination: if each voter abstains so that their optimal weight divided by the number of votes they cast equals the same common number R, the realized voting shares become proportional to the optimal weights. That makes partial abstention epistemically superior, in this model, to transfer delegation in any form, including liquid democracy, which requires voters to know other voters' competences and suffers from over-delegation. The paper also shows that more direct participation helps only when the voting rule already uses optimal weights; outside that environment, adding low-competence or correlated voters can sharply reduce the probability of a correct outcome.","feed_headline":"Partial abstention beats transfer delegation in DAO votes","feed_subtitle":"When all voters abstain to one weight-to-vote ratio, the vote reproduces optimal weights with little coordination.","key_machinery":"The load-bearing device is the Optimal Weighting Theorem: for independent binary voters, the decision rule that maximizes the probability of a correct collective choice weights each voter by w_i = ln(p_i/(1-p_i)), the log odds of the voter's competence. The paper's new contribution is the ratio-symmetric abstention rule w_i/t*_i = R: because every voter's exercised votes are scaled to their optimal weight by the same constant, the realized vote shares are exactly proportional to the optimal weights and the aggregate decision is identical to the social-planner optimum. A second device, the canonical independent-signals decomposition, extends the same arithmetic to correlated voters by treating shared signals as a single weighted signal discounted by its audience count.","core_discovery":"The paper's central claim is Proposition 3.2: with independent voter competencies and at least one voter with p_j > 0.5, if every voter partially abstains so that w_i/t*_i = R, where w_i = ln(p_i/(1-p_i)) is the voter's optimal weight and R is the supremum of w_i/t_i over voters, then the probability of a correct collective decision under weighted voting is maximized. The argument is an identity: the abstention ratios make effective weights proportional to optimal weights by a common factor, so the two voting rules are decision-equivalent. A corollary is that the optimum is reached with only common knowledge of the single number R, or any sufficiently large substitute, rather than a social planner or knowledge of other voters' competences. In the dependent-competencies case the same logic applies to independent signals: each voter discounts each signal by the number of voters who share it, restoring optimal weighting under the paper's information-decomposition and signal-numeracy assumptions. The paper therefore asserts that the main obstacle to optimal decentralized voting is not aggregation but the accuracy of each voter's self-assessed competence.","pith_inferences":["The same ratio identity applies to any weighted committee or expert panel with competence estimates, so the result is not tied to blockchain tokens; a medical or scientific advisory body could adopt partial abstention with a publicly agreed R.","A practical DAO could implement the rule in code: voters report p_i, the contract computes w_i and instructs each wallet to cast w_i/R of its vote weight, turning the paper's coordination condition into an implementable default.","If self-reported p_i are verifiable ex post against voting records, the mechanism could be combined with reputation or slashing to discipline miscalibration, which the paper itself identifies as the fragile point.","The flooding danger suggests that turnout policy should be signal-based: recruiting voters who bring independent information helps, while blanket participation campaigns can actively reduce decision accuracy."],"forward_implications":["Under independent competencies, a DAO can attain the best possible epistemic outcome using only one public reference number and each voter's own competence estimate, with no delegate network, social planner, or knowledge about other voters.","Any governance method built on indivisible votes, including ordinary majority voting and liquid democracy with all-or-nothing delegation, is epistemically dominated by partial abstention in the environments the paper models.","A policy of increasing direct participation is guaranteed to do no epistemic harm only inside the optimal epistemic environment; otherwise new voters can flood the decision with low-weight or correlated signals and push the probability of correctness below 0.5.","Transfer delegation, even in multi-step form, requires delegative competence, knowing other voters' optimal weights, and is vulnerable to over-delegation; the paper gives only a narrow connected-network condition under which three-step transfer delegation can reach the optimum.","With dependent competencies, optimal decentralized outcomes require each voter to know how many others share each signal, so dependency tracking, not abstention arithmetic, is the binding constraint."],"supporting_citations":[{"why":"Supplies the Optimal Weighting Theorem and Lemma 4.2, the baseline any governance method must match.","marker":"Nitzan and Paroush (1982)"},{"why":"Defines the jury-theorem ideal of increasing reliability that the two epistemic tests generalize.","marker":"de Condorcet (1785)"},{"why":"Establishes abstention as a form of delegation to superior information, the conceptual base of Observation 3.1.","marker":"Feddersen and Pesendorfer (1996)"},{"why":"Provides the independent-signal decomposition that the dependent-competencies version of partial abstention relies on.","marker":"Dietrich and Spiekermann (2024)"},{"why":"Gives the sufficient condition for added voters to improve majority outcomes, used in the participation analysis.","marker":"Ben-Yashar and Nitzan (2017)"},{"why":"Shows local delegation mechanisms cannot bound losses and motivates over-delegation concerns in liquid democracy.","marker":"Kahng et al. (2021)"},{"why":"Experiments compare liquid democracy with majority voting and abstention and document over-delegation, the empirical foil for partial abstention.","marker":"Mooers et al. (2024)"},{"why":"Underwrites the Condorcet AI-agent supplement by showing prediction markets can approximate optimal-weight decisions.","marker":"Airiau et al. (2024)"}],"fun_headline_variants":["Partial abstention wins over delegation in DAOs","DAO voting: abstain partially for optimal outcomes","Epistemic edge: partial abstention in decentralized votes","How to vote wisely in DAOs: partial abstention","The secret to DAO voting: partial abstention"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Voters can compute an accurate probability p_i that their own information leads to the correct choice, and the optimality of the abstention rule collapses if those self-assessments are miscalibrated.","fun_headline_variants_meta":{"raw":{"variants":["Partial abstention wins over delegation in DAOs","DAO voting: abstain partially for optimal outcomes","Epistemic edge: partial abstention in decentralized votes","How to vote wisely in DAOs: partial abstention","The secret to DAO voting: partial abstention"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000244,"raw_usage":{"total_tokens":1547,"prompt_tokens":973,"completion_tokens":574,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":589,"completion_tokens_details":{"reasoning_tokens":498}},"tokens_in":589,"tokens_out":574,"duration_ms":5989,"temperature":1.0,"reasoning_tokens":498,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T23:36:29.682738+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a binary-choice task with known correct answers, elicit each participant's self-assessed p_i, apply the abstention rule w_i/t*_i = R, and compare collective accuracy against simple majority and the social-planner optimum. The claim is falsified if the abstention rule underperforms a naive rule whenever self-assessed p_i are miscalibrated.","supporting_citations":[{"cited_title":"Con- dorcet Markets,","cited_arxiv_id":null,"evidence_quote":"Underwrites the Condorcet AI-agent supplement by showing prediction markets can approximate optimal-weight decisions."}],"review_version":1}