{"id":"933036c7-f89b-4ec7-ab11-04733bdd3695","arxiv_id":"2608.01228","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For ℓ_t utilities, individual fair share guarantees are computable and compatible with Pareto efficiency; for ℓ1 and ℓ2 utilities, strategyproofness, efficiency, and even the weakest single-minded fairness cannot all hold whenever there are at least three alternatives.","lead":"This paper studies how to fairly combine people's preferred budget splits into one shared budget split, and proves which fair outcomes can be guaranteed at all. It shows that stronger individual fairness can coexist with efficiency for many preference models, but also proves that for three or more options, no rule can be simultaneously fair, efficient, and manipulation-proof.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The ℓ2 impossibility rests on an unsupported 'hence' in Lemma C.5; Pareto efficiency alone allows a continuum of possible outcomes.","rationale":"The reader identified Proposition 5.1 as the weakest assumption, but that proposition appears correct: the metric-projection argument validly extends the Durier-Michelot characterization to the simplex. The more serious issue is in Lemma C.5, where the proof jumps from 'outcome cannot be p_N1' to 'outcome must be p'_N1' without using strategyproofness. This is not a mere typo, because the intermediate points of the segment are Pareto efficient and are not excluded by any statement in the appendix. The gap is repairable—one can likely derive a contradiction by moving N2 to the current outcome and applying Lemma C.4—but as written the central ℓ2 results depend on an unproved claim. Theorem 4.3 for ℓ1 utilities appears well supported by the chain-of-profiles argument. The paper's overall contribution is strong, so the appropriate change is to make acceptance conditional on completing or correcting the proof of Lemma C.5.","tokens_in":30115,"tokens_out":33273,"duration_ms":288958,"concrete_test":"Provide a complete proof of the sentence 'Hence, the outcome must be p'_N1' in Lemma C.5, using only the definitions and lemmas already established. The proof must handle the cases where the efficient outcome is p'_N1, q_epsilon, or an interior point of conv(q_epsilon, p'_N1). If the only way to finish the step is to add an assumption not present in the paper, such as continuity of the rule, then Theorem 5.3 should be marked conditional on that assumption.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Section 5's oligarchy/dictatorship argument hinges on Lemma C.5. The critical step asserts that after all agents in N2 have moved their peaks from p_N1 to q_epsilon = (1-epsilon)p_N1 + epsilon p'_N1, 'By efficiency, the new outcome cannot be p_N1, as it is outside the convex hull of peaks. Hence, the outcome must be p'_N1.' The first assertion is true. The second does not follow from efficiency alone: Proposition 5.1 only places the outcome in conv(q_epsilon, p'_N1), i.e. on the segment between q_epsilon and p'_N1, and every interior point of that segment is Pareto efficient. Strategyproofness is not invoked in the quoted step. To make the step valid one must give an additional argument—for example, that if the outcome were any z different from p'_N1, moving N2's peaks to z would make N2 decisive at z and contradict the already-established decisiveness of N1 at p_N1. Such an argument is not present in the text, and the special cases z = p'_N1 and z = q_epsilon require care. Since Lemma C.5 is the bridge from local decisiveness to global oligarchy, and Lemma 5.2 and Theorem 5.3 depend on it, the ℓ2 impossibility and the n = 2 dictatorship characterization are not fully established by the written proof.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies budget aggregation, where each agent reports a preferred distribution over m alternatives and a rule returns a collective distribution. It defines two individual-fair-share notions (IFS1 and IFS2), a weaker positive-share (PS) notion, and a single-minded weakening (single-minded PS). For general ℓ_t utilities, it gives polynomial-time algorithms for computing the IFS values and proves that a distribution can simultaneously satisfy IFS2 and Pareto efficiency. For ℓ_1 utilities it derives closed-form IFS formulas, proves an impossibility of strategyproofness, Pareto efficiency, and single-minded PS for n,m ≥ 3, and gives positive rules for m=2 and for (n,m)=(2,3). For ℓ_2 utilities it proves a similar impossibility for all n≥2, m≥3 and characterizes the two-agent case as dictatorship via an oligarchy lemma.","tokens_in":1206,"tokens_out":1735,"duration_ms":256879,"significance":"If the results are correct, they materially strengthen earlier impossibility results by replacing single-minded proportionality with the much weaker single-minded positive share, and they provide a clean oligarchy/dictatorship characterization for ℓ_2 utilities. The paper's constructive results for IFS2, especially the water-filling algorithms and the existence of efficient distributions satisfying IFS2, are valuable and are supported by explicit proofs. The ℓ_1 impossibility proof is self-contained and appears sound. The main obstacle is the proof of Lemma C.5 in the ℓ_2 section, which is load-bearing for the Section 5 conclusions and currently contains an unjustified step; the gap looks repairable, but it must be fixed before the paper's central ℓ_2 claims are fully established.","major_comments":[{"comment":"The step reading 'By efficiency, the new outcome cannot be p_N1, as it is outside the convex hull of peaks. Hence, the outcome must be p'_N1' is not justified. After moving the peaks of agents in N2 to q_epsilon = (1-epsilon)p_N1 + epsilon p'_N1, Proposition 5.1 only implies that the new outcome lies in conv(q_epsilon, p'_N1), i.e., on the segment between q_epsilon and p'_N1, and every interior point of that segment is Pareto efficient. Strategyproofness is not invoked in this sentence, so efficiency alone does not force the outcome to be p'_N1. This is a load-bearing point: Lemma C.5 is the bridge from local decisiveness to the global oligarchy in Lemma 5.2, and Theorem 5.3 depends on Lemma 5.2. The gap appears repairable: if the outcome were some z different from p'_N1, one could move all agents in N2 to z, apply strategyproofness to keep the outcome at z, and then use Lemma C.4 to make N2 decisive at z, contradicting the already-established decisiveness of N1 at p_N1; the boundary cases z=q_epsilon and z=p_N1 need separate care. This argument, however, is not present in the manuscript and must be added.","section":"Appendix C, Lemma C.5"}],"minor_comments":[{"comment":"The assertion 'By strategyproofness, f(P')=f(P)' in the chain of profiles is too terse. Strategyproofness alone does not force the outcome to stay fixed when a voter changes her report; the argument needs the induction that the outcome before each move is already f(P), so that an agent with true peak f(P) could profitably deviate if the outcome changed. This is a valid implicit argument, but it should be spelled out.","section":"Appendix C, Lemma C.4"},{"comment":"The sentence 'This impossibility shows that every unanimous and (strongly) group-strategyproof rule violates positive share' should be made more precise: the theorem establishes a violation of single-minded positive share, and one additional sentence is needed to explain that a rule satisfying positive share would in particular satisfy single-minded positive share.","section":"Section 4.2, discussion after Theorem 4.3"},{"comment":"The abbreviation IFS0 appears in Table 1 and in the related-work discussion before it is formally defined in Section 4; consider moving or anticipating the definition so that all uses of the term are defined at first appearance.","section":"Section 4, paragraph introducing IFS0"}],"recommendation":"major_revision","confidential_remarks":"The paper is well within the scope of the journal and the contributions are substantial. The only substantive concern is the gap in Lemma C.5, which affects the completeness of the ℓ_2 impossibility and the two-agent dictatorship theorem; the existing lemmas suggest a local repair, so I do not see grounds for rejection if the proof is corrected."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague, here's my take. The paper is a serious theory contribution, but it's not all of a piece. The Sections 3 and 4 material is genuinely good. Introducing IFS2 and computing IFS values in strongly polynomial time for all ℓ_t utilities via water-filling is clean; the existence of distributions meeting IFS2 plus Pareto efficiency (Cor 3.10) and the explicit formulas for ℓ1 (Thms 4.1, 4.2) are real results. Theorem 4.3, the impossibility for ℓ1 when n,m≥3 with only single-minded positive share, is a strong strengthening of Brandt et al. and the proof, while long, is coherent. The input-dependent phantom-order rule for (2,3) is an interesting mechanism that deserves attention.\n\nThe problem is Section 5. Lemma C.5 has a load-bearing gap. After moving N2's peaks to q_epsilon, the text says \"By efficiency, the new outcome cannot be p_N1, as it is outside the convex hull of peaks. Hence, the outcome must be p'_N1.\" That 'hence' does not follow. Proposition 5.1 only says the outcome is in conv(q_epsilon, p'_N1) — the whole segment. Every interior point of that segment is Pareto efficient, and strategyproofness is not invoked at this step. The quoted argument doesn't rule out an interior point. Without an extra argument — e.g., showing that an interior outcome z would let N2 become decisive at z and contradict N1's decisiveness at p_N1 — the oligarchy lemma, Lemma 5.2, and Theorem 5.3 are not established. That's a serious gap, not a cosmetic one, because Section 5 is billed as the ℓ2 analogue of the main impossibility.\n\nThere are smaller issues: Prop 5.1's extension of Durier-Michelot is asserted without proof in the main text; the appendix's projection argument is compressed about boundary points. And Lemma A.1's Nash equilibrium characterization is informal, though I don't think it's wrong. None of these approach the C.5 problem.\n\nIn short: Sections 3 and 4 deserve a careful read and are citable. Section 5 is not there yet. If the authors can fill the C.5 gap, the paper is strong; if not, the ℓ2 claims should be withdrawn or weakened. I would send it to peer review, but with a request for a complete proof of Lemma C.5 before acceptance.","headline":"Strong on ℓ1 and computation; Section 5's ℓ2 oligarchy proof has an unjustified step that knocks out the headline ℓ2 impossibility as written.","tokens_in":30949,"tokens_out":7252,"would_cite":true,"duration_ms":58974,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91B14","91B32"],"pacs":[],"model":"deepseek-v4-flash","headline":"In budget aggregation with at least three alternatives, no rule can be simultaneously strategyproof, Pareto efficient, and fair in even the weakest single-minded sense; the same holds for both $\\ell_1$ and $\\ell_2$ disutility models.","keywords":["budget aggregation","individual fair share","strategyproofness","Pareto efficiency","single-minded positive share","ℓt utilities","moving-phantom rules","social choice theory"],"falsifier":"Take $m=3$, two agents with peaks at $(1,0,0)$ and $(0,1,0)$ under $\\ell_2$ utilities. Compute whether any distribution outside the segment between these peaks Pareto dominates a distribution on the segment; finding one would refute Proposition 5.1 and the $\\ell_2$ dictatorship theorem. For $\\ell_1$, an exhaustive computer search over a fine grid of profiles for $n=m=3$ looking for any rule that is strategyproof, Pareto efficient, and gives strictly positive overlap to each single-minded agent would either locate such a rule (refuting Theorem 4.3) or corroborate it.","tokens_in":29844,"feed_emoji":"🗳️","tokens_out":11699,"duration_ms":99381,"temperature":0.7,"pith_summary":"This paper asks what fairness one can require when a group aggregates individual preferred distributions of a fixed budget. It introduces two versions of individual fair share and a much weaker requirement, single-minded positive share, and proves that this weakest requirement cannot be combined with strategyproofness and Pareto efficiency once there are at least three alternatives. The incompatibility holds for both $\\ell_1$ (Manhattan) and $\\ell_2$ (Euclidean) disutilities. Against that, the paper shows that stronger fair-share guarantees can always be satisfied together with Pareto efficiency for any $\\ell_t$ utility, with polynomial-time computation, and that in two small cases they can even be combined with strategyproofness. The reason this matters is that even the minimal protection of a single-minded minority forces a mechanism designer to give up either truthfulness or efficiency.","feed_headline":"No budget rule can be fair, truthful, and efficient","feed_subtitle":"With three or more options, even a minimal single-minded fairness guarantee forces a choice between truthfulness and efficiency.","key_machinery":"For the negative $\\ell_2$ results, the load-bearing object is the convex hull of the reported peaks: Proposition 5.1 identifies Pareto efficient outcomes with points in $\\operatorname{conv}(p_1,\\ldots,p_n)$, and a local-dictatorship lemma then forces any strategyproof efficient rule to be an oligarchy—once a coalition's peak coincides with the outcome in a two-peak profile, it must dictate whenever its members agree. For $\\ell_1$, the analogous mechanism is the influence of an isolated single-minded agent, measured by her overlap utility; strategyproofness requires this influence to survive along a chain of intermediate profiles, while two agents with incompatible peaks require it simultaneously, producing the contradiction. On the constructive side, a water-filling procedure—each agent spends her $1/n$ share on the alternatives with the largest current deficit relative to her peak—provides the IFS values and a pure Nash equilibrium, from which IFS2-satisfying Pareto efficient distributions are derived.","core_discovery":"The paper's central discovery is an incompatibility: for $\\ell_1$ utilities with $n,m\\ge 3$, no aggregation rule can satisfy strategyproofness, Pareto efficiency, and single-minded positive share—the requirement that in profiles where every agent's peak is a single alternative, each agent receives strictly more than her worst possible utility. For $\\ell_2$ utilities the same incompatibility holds for any $n\\ge2$ and $m\\ge3$, and for two agents it sharpens to a full characterization: every strategyproof and Pareto efficient rule is a dictatorship. Balanced against this, the paper shows that for $\\ell_t$ utilities with $t\\ge1$, distributions giving every agent the stronger individual-fair-share guarantees IFS1 and IFS2 always exist together with Pareto efficiency and can be computed in polynomial time, and that in the small cases $m=2$ or $(n,m)=(2,3)$ with $\\ell_1$, fully truthful, efficient, and IFS2-satisfying rules exist.","pith_inferences":["Beyond the paper, the convex-hull/dictatorship route used for $\\ell_2$ suggests testing the same impossibility for other strictly convex $\\ell_t$ metrics with $1<t<\\infty$; a direct test would be to rerun the two-peak local-dictatorship argument for $\\ell_3$.","A practical consequence the authors leave implicit is that in participatory budgeting, any mechanism that is both strategyproof and Pareto efficient will systematically ignore the preferences of a single-minded minority, so designers must either accept bounded manipulation or weaken one of the two axioms.","The input-dependent phantom-order rule suggests a general design pattern—letting the tie-breaking order of phantom movements depend on the reported peaks—that could be explored for $m>3$ alternatives."],"forward_implications":["For $\\ell_1$ and $\\ell_2$, any mechanism designer who wants truthful reporting and Pareto efficiency must give up even the weakest individual fairness once $m\\ge3$; no rule can have all three.","For $m=2$, truthful efficient aggregation with strong individual fairness is possible for all $n$ via the uniform phantom rule.","For $(n,m)=(2,3)$ under $\\ell_1$, the input-dependent phantom-order rule shows that relaxing neutrality allows strategyproofness, Pareto efficiency, anonymity, continuity, and IFS2 to coexist.","For $\\ell_2$ with $n=2$, the dictatorship characterization makes the impossibility transparent: the only truthful efficient rules are those that let one agent decide.","The polynomial-time construction of IFS2-satisfying efficient distributions means fairness and efficiency alone are always compatible for any $\\ell_t$; only adding strategyproofness creates the deadlock."],"supporting_citations":[{"why":"Introduces moving-phantom rules and single-minded proportionality, the benchmark family that the negative theorems strengthen and that supplies the m=2 uniform phantom rule.","marker":"Freeman et al. (2021)"},{"why":"Established the prior $\\ell_1$ impossibility with single-minded proportionality that Theorem 4.3 replaces with the weaker single-minded positive share.","marker":"Brandt et al. (2026)"},{"why":"Supplies the critical profile and continuity argument for the m>n impossibility with IFS0.","marker":"de Berg et al. (2024)"},{"why":"Gives the Hilbert-space result that efficient points are the closed convex hull of peaks, which Proposition 5.1 extends to the budget simplex.","marker":"Durier and Michelot (1986)"},{"why":"Introduced IFS1 and the budget-aggregation game; its open question about polynomial-time equilibria is resolved by Proposition 3.7.","marker":"Becker et al. (2026)"},{"why":"Origin of individual fair share and positive share for dichotomous preferences, adapted here to distance-based utilities.","marker":"Bogomolnaia, Moulin, and Stong (2005)"},{"why":"Provides the phantom-voter characterization of strategyproof one-dimensional rules used for the uniform phantom rule.","marker":"Moulin (1980)"}],"fun_headline_variants":["Fair, truthful, efficient: pick two in budget aggregation","Budget aggregation trilemma: fairness, truth, efficiency","Three options doom fair, truthful, efficient budget rules","No fair, honest, and efficient budget rule for 3+ options"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The $\\ell_2$ result leans on the claim that in Euclidean space an efficient outcome must lie inside the convex hull of the reported peaks, extended to the budget simplex; if a boundary point of the simplex were an exception, the chain of lemmas that produces a dictatorship would collapse.","fun_headline_variants_meta":{"raw":{"variants":["Fair, truthful, efficient: pick two in budget aggregation","Budget aggregation trilemma: fairness, truth, efficiency","Three options doom fair, truthful, efficient budget rules","No fair, honest, and efficient budget rule for 3+ options"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000754,"raw_usage":{"total_tokens":3329,"prompt_tokens":895,"completion_tokens":2434,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":511,"completion_tokens_details":{"reasoning_tokens":2366}},"tokens_in":511,"tokens_out":2434,"duration_ms":16791,"temperature":1.0,"reasoning_tokens":2366,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T15:11:35.133865+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take $m=3$, two agents with peaks at $(1,0,0)$ and $(0,1,0)$ under $\\ell_2$ utilities. Compute whether any distribution outside the segment between these peaks Pareto dominates a distribution on the segment; finding one would refute Proposition 5.1 and the $\\ell_2$ dictatorship theorem. For $\\ell_1$, an exhaustive computer search over a fine grid of profiles for $n=m=3$ looking for any rule that is strategyproof, Pareto efficient, and gives strictly positive overlap to each single-minded agent would either locate such a rule (refuting Theorem 4.3) or corroborate it.","supporting_citations":[],"review_version":2}