{"id":"bebd69e4-e2b1-4d2a-8078-faf1cbba078e","arxiv_id":"2607.21662","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A margin-based moment estimator recovers the three coupling parameters of a two-group Curie-Weiss voting model with asymptotic normality in the weak-interaction regime; in the strong-interaction regime only the magnetization point, not the couplings, is identifiable.","lead":"This paper builds a fast estimator for the interaction strengths in a two-group Curie-Weiss model of voting, using only the vote totals of each group from a sample of elections, and proves consistency and asymptotic normality in the weak-coupling regime. The practical takeaway is a fitting recipe plus a warning: when group interactions are strong, the coupling parameters cannot be recovered from the same statistics.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proposition 8 – the unproved low-temperature uniqueness and correlation limit – is the load-bearing block; Theorem 11's N→∞ consistency is only a restatement of it.","rationale":"The reader's weakest_assumption already identifies Proposition 8, and I agree. Reading the proof structure of Theorem 11, Proposition 8 is the only non-standard input. The high-temperature part is believable but still not proven in-text; the low-temperature part, with its uniqueness and sign-flip mixture claim, is exactly where a subtle failure could occur. The numerical simulations do not independently verify Proposition 8 because they are for two specific matrices and report no error bars or convergence plots. The other concerns (CLT arrow, delta-method Jacobian, table mismatch) are real but secondary; they affect the precision of Theorem 11.3 and the presentation, not the central consistency theorem. Therefore the appropriate verdict is unchanged: conditional acceptance pending a proof or precise citation of Proposition 8. I am not asserting the proposition is false; the concern is that the central claim is unsupported by the manuscript as submitted.","tokens_in":11327,"tokens_out":12097,"duration_ms":119462,"concrete_test":"Independently derive the low-temperature clause of Proposition 8 by Laplace's method: write the moment generating function of (S1/√N1, S2/√N2) as an integral over the free energy f(m) = 1/2 m^T J m - (1/2)Σ_λ[(1+mλ)log(1+mλ)+(1-mλ)log(1-mλ)] and show that the set of global minima of f is exactly { (m1,m2), (-m1,-m2) } with m1>0 and sign(m2)=sign(J12) for every J>0 with I-J not PSD and J12≠0. If the derivation requires an extra hypothesis (e.g., J12>0, or an upper bound on |J12|), Proposition 8 as stated is overbroad and the low-temperature half of Theorem 11 must be restricted accordingly. Alternatively, run a dense grid over (J11,J22,J12) satisfying the admissibility conditions, solve the stationary equations numerically, and check the global-minimum structure; any counterexample would refute the proposition.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The asymptotic content of Theorem 11 is entirely inherited from Proposition 8. Theorem 11.1 and 11.2 (high temperature) are the WLLN plus a continuity argument around Definition 7; once Proposition 8 is granted, the rest is standard. In low temperature, Theorem 11.2's second claim is literally Proposition 8's second sentence. Proposition 8 is stated without proof, with footnote 1 pointing to 'articles' that are not accessible in the manuscript. The high-temperature susceptibility limit E[SλSν]/sqrt(NλNν) -> (I-J)^{-1} is a standard mean-field result, but it is not derived here. The low-temperature clause is more delicate: it asserts a unique (m1,m2) with m1>0, m2≠0 and the convergence of second moments to mλmν. That requires both uniqueness of the global minimizer of the mean-field free energy up to the global spin flip and concentration of the two-group Gibbs measure on those two points. The paper's observation that 'the sign of m2 is the same as that of J12' does not establish uniqueness, and the mixture structure (the signs of the two group magnetizations flip together) is simply assumed. If for any admissible J (positive definite, I-J not PSD, J12≠0) the free energy has more than one pair of global minima or a global minimum with m2=0, both low-temperature parts of Theorem 11.2 and the low-temperature identifiability conclusions collapse. The secondary issues (the p→ vs d→ in Theorem 11.3, unverified det(Δ(μ))≠0, table/text mismatch) do not affect the main consistency claim, but Proposition 8 is the hinge.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies estimation for a two-group Curie-Weiss (block-spin Ising) model with interacting groups, where the full Hamiltonian has coupling matrix J, and one observes n i.i.d. complete ballot configurations. The authors propose a plug-in estimator: in the high-temperature regime (I-J > 0), J-hat_{N1,N2,n} = I - N T(x)^{-1} N, where T is the empirical matrix of squared/ cross group sums; in the low-temperature regime, m-hat = N^{-1} N^{-1} T N^{-1} N^{-1} estimates the matrix of magnetization products. The main theoretical result, Theorem 11, asserts: (1) J-hat converges in probability to a finite-population quantity J-tilde_{N1,N2}; (2) as N1,N2 -> infinity, J-tilde_{N1,N2} -> J in the high-temperature case and m-tilde_{N1,N2} -> [[m1^2, m1 m2],[m1 m2, m2^2]] in the low-temperature case; and (3) a CLT with explicit covariance. The proof of Theorem 11 is largely a combination of the weak law of large numbers, the delta method, and Proposition 8, which supplies the asymptotic behavior of the model moments E[S_lambda S_nu] in both regimes. The paper also includes simulations for one high-temperature and one low-temperature coupling matrix.","tokens_in":11538,"tokens_out":3583,"duration_ms":38299,"significance":"If the results are correct, the paper provides a computationally simple, closed-form estimator for the coupling parameters of a two-group Curie-Weiss model, avoiding the partition-function evaluation needed for maximum likelihood. This is a useful contribution to the growing literature on inverse problems for mean-field spin systems and social-choice applications. The high-temperature estimator and its CLT are natural and the algebraic steps in Section 6 are mostly standard. However, the scientific weight of the paper rests almost entirely on Proposition 8, which is stated without proof or accessible reference. In particular, the low-temperature uniqueness and the convergence of second moments to m_lambda m_nu are asserted rather than established. Because Theorem 11.2 is essentially a restatement of Proposition 8, the paper's central consistency claims are unproven unless Proposition 8 is supplied. The simulations illustrate behavior for two specific matrices but do not substitute for the missing proof.","major_comments":[{"comment":"Proposition 8 is the load-bearing block of the paper: Theorem 11.2 follows directly from it, and Theorem 11.1 is only the WLLN plus Definition 7. Yet the proposition is stated without proof, and the only justification is footnote 1, which refers to 'articles' that are not accessible. For the high-temperature part, the limit E[S_lambda S_nu]/sqrt(N_lambda N_nu) -> (I-J)^{-1}_{lambda nu} is standard but not derived. For the low-temperature part, the assertion of a unique (m1,m2) with m1>0 and m2≠0, and the convergence of E[S_lambda S_nu]/(N_lambda N_nu) to m_lambda m_nu, implicitly assumes both uniqueness of the global minimizer modulo spin-flip and concentration of the two-group measure on the two flipped configurations. The remark that 'the sign of m2 is the same as that of J12' does not establish either. If, for some admissible J with I-J not PSD, the mean-field free energy has more tha","section":"Section 3, Proposition 8 and footnote 1"},{"comment":"The CLT statements use the notation \\sqrt{n}(\\hat J - \\tilde J) \\xrightarrow{p} N(0,C) and \\sqrt{n}(\\hat m - \\tilde m) \\xrightarrow{p} N(0,D). Convergence in probability to a normal distribution is not the intended statement; the correct mode is convergence in distribution, \\xrightarrow{d}. This appears in both high- and low-temperature statements of Theorem 11.3 and should be corrected. Additionally, for the high-temperature CLT, the proof applies Theorem 14 but never verifies its assumption det(Delta(mu)) \\neq 0. The paper asserts that C is non-singular, but that requires checking the determinant. If det(Delta(\\tilde J)) = 0 for some admissible J, the delta-method CLT as stated is invalid. This is a checkable linear-algebra condition and should be either proven or the theorem weakened accordingly.","section":"Section 5.1, Theorem 11.3 and proof in Section 6.3"},{"comment":"Definition 7 defines J-tilde_{N1,N2} implicitly by (I - J-tilde)^{-1} = E[diag-style matrix of scaled second moments]. Consequently, Theorem 11.1, asserting \\hat J_{N1,N2,n} \\xrightarrow{p} J-tilde_{N1,N2}, is an immediate consequence of the WLLN and the continuity of matrix inversion; it is true by construction. This should be stated explicitly to avoid giving the impression that the main content of Theorem 11.1 is a statistical consistency result. The substantive statistical claim is the limit J-tilde -> J as N1,N2 -> infinity, which is exactly Proposition 8. The paper's presentation should separate these two layers, especially since the reader may otherwise mistake Definition 7 for a model assumption rather than a definition.","section":"Section 3 and Definition 7"}],"minor_comments":[{"comment":"The text says the estimator for (m1^2, m2^2) was within tolerance 100% of the true value, but Table 2 reports a statistic named '(T(x)_{1,1}, T(x)_{2,2})' within tolerance. The table label does not match the estimator described in the text; clarify whether the table reports T or the m-hat-based estimator.","section":"Section 5.2.2, Table 2"},{"comment":"The theorem states \\Upsilon > 0 and then uses \\Sigma in equation (9). This is a typo; also the domain is written as D \\subset \\mathbb{R} but should be D \\subset \\mathbb{R}^d for a d-dimensional delta method.","section":"Appendix, Theorem 14"},{"comment":"The sentence 'The estimate \\hat J_{N1,N2,n}(x) lay within a tolerance of 0.1 of the true coupling matrix in (4) 20% of the time' is inconsistent with Table 1, which says 27% for the same row. One of the two numbers is wrong.","section":"Section 5.2.1"},{"comment":"The notation (\\hat m)^2_1 and (\\hat m)^2_2 is misleading because \\hat m is a 2x2 matrix. Use explicit entries such as ((\\hat m)_{1,1})^2, ((\\hat m)_{2,2})^2, and sign((\\hat m)_{1,2}).","section":"Section 5.2.2"}],"recommendation":"major_revision","confidential_remarks":"The manuscript's central claim is conditional on Proposition 8, which is unproved and cited only to unavailable 'articles'. This is not a presentation issue but a missing proof of the paper's main ingredient. The high-temperature part is standard and can likely be supplied; the low-temperature uniqueness and second-moment convergence require more delicate analysis. If the authors can provide a complete proof of Proposition 8 (or a precise citation to a theorem that covers exactly this model), the paper could be suitable for publication. I also recommend checking the determinant condition for the delta method and correcting the CLT convergence mode."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The real step beyond the authors' previous work is the estimator for the cross-group coupling J12 in a two-group Curie-Weiss model. That is a legitimate gap in the literature, and the plug-in estimator (I - N T^{-1} N) is natural and computationally cheap. The paper is clearly written, and the high-temperature consistency follows transparently from the WLLN plus continuity once the susceptibility limit in Proposition 8 is granted. I agree with the reader that Proposition 8 is the load-bearing block. It is stated without proof, with a footnote pointing to 'articles' that are not accessible. In the low-temperature regime it assumes both a unique minimizer (m1,m2) with m1>0, m2≠0 and the symmetric mixture structure (both groups flip together). Those are substantive facts about the Gibbs measure, not trivial consequences of the model. If they fail for admissible J, Theorem 11.2's low-temperature claim and the identifiability discussion collapse. The paper does not settle that. On the secondary issues: Theorem 11.3 uses 'p' for convergence in distribution, which is a notation error; non-singularity of C and det(Δ(μ))≠0 are asserted but not checked. The simulations have a real inconsistency—text says 20% within tolerance for n=20, table says 27%—and the low-temperature estimator description is unclear about how the sign of m2 is recovered. No code or data is provided. None of these are fatal on their own; they are fixable. But they add to the sense that the paper was submitted before the details were fully tightened. Who is this for? Researchers working on inverse problems for mean-field spin models, especially those who want a cheap alternative to MLE for multi-group data. It deserves a serious referee: the central idea is worth engaging with, and a referee could demand a proof or precise reference for Proposition 8, plus the small corrections. I would not cite the low-temperature results until the proof appears, but I would send this to peer review.","headline":"A plausible but hinge-on-unproved Proposition 8; worth a referee, not a quick accept.","tokens_in":12232,"tokens_out":1724,"would_cite":false,"duration_ms":18464,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["62F10","82B20","60F05","91B12"],"pacs":[],"model":"deepseek-v4-flash","headline":"For a two-group Curie-Weiss model, a plug-in estimator built from the empirical second moments of group voting margins consistently recovers the coupling matrix in the high-temperature regime and the concentration point in the low-temperatu","keywords":["Curie-Weiss model","multi-group mean-field model","statistical mechanics","Gibbs measures","large population approximation","voting behavior","coupling parameter estimation"],"falsifier":"Fix a low-temperature, positive-definite J with negative cross-coupling (for example J11=2.0, J22=1.8, J12=-1.0), simulate the two-group Curie-Weiss model for large N1=N2=N, and compute E[S1S2]/(N1N2). If the limit does not approach m1m2 with sign(m2)=sign(J12), or if two distinct (m1,m2) points both reproduce the observed product limits, Proposition 8 is false and the theorem's low-temperature branch fails.","tokens_in":11040,"feed_emoji":"🗳️","tokens_out":8826,"duration_ms":63411,"temperature":0.7,"pith_summary":"The paper studies statistical reconstruction of a two-group Curie-Weiss (mean-field) model, in which voters in two groups influence each other both within and across groups. It proposes plug-in estimators built from the empirical second moments of each group's voting margin: the high-temperature estimator is I - N T^{-1} N, and the low-temperature estimator is N^{-1}N^{-1} T N^{-1}N^{-1}. The main theorem shows that as the number of ballot samples grows, the high-temperature estimator converges in probability to a finite-population target J-tilde, which itself converges to the true coupling matrix J as group sizes grow, and that a scaled version converges to a centered Gaussian with an explicit covariance. In the low-temperature regime the estimator converges to the product m_lambda m_nu of the coordinates of the model's concentration point, providing information about coupling signs and magnitudes. The paper thus supplies a computationally simple alternative to maximum likelihood, which requires evaluating an intractable partition function.","feed_headline":"Ballot margins recover two-group Curie-Weiss interactions","feed_subtitle":"Sample second moments of two groups' vote margins estimate all coupling parameters, with an explicit Gaussian error.","key_machinery":"Proposition 8, deferred to prior work for proof, supplies the asymptotic second-moment structure: in the high-temperature regime the normalized margin covariances approach the inverse of I-J, and in the low-temperature regime they approach the rank-one product m_lambda m_nu of the unique concentration point (m1,m2). The plug-in estimators are built from T, the empirical matrix of those second moments; the high-temperature estimator applies the inverse map I - N T^{-1} N, and the low-temperature estimator applies the scaling N^{-1}N^{-1} T N^{-1}N^{-1}. The delta method then turns the multivariate CLT for T into the stated CLTs for the estimators.","core_discovery":"The central object is the statistic T, the sample average of the outer product of the two groups' voting margins. In the high-temperature regime the plug-in J-hat = I - N T^{-1} N converges to J-tilde_{N1,N2} as the number of ballots n grows, and J-tilde converges to J as populations N1,N2 grow; scaled error sqrt(n)(J-hat - J-tilde) converges to a centered Gaussian with covariance C given explicitly by a delta-method sandwich formula. In the low-temperature regime, m-hat = N^{-1}N^{-1} T N^{-1}N^{-1} converges to the rank-one matrix [[m1^2, m1m2],[m1m2, m2^2]] for the model's unique concentration point, identifying the sign of the cross-group coupling because sign(m2) = sign(J12). The engine","pith_inferences":["Editorial inference: Because the low-temperature limit is the rank-one matrix m m^T, the estimator identifies m only through products; if the model were modified to allow asymmetric coupling, this rank-one structure would break and identifiability would change.","Editorial inference: The theorem covers the disjoint high- and low-temperature regimes; at the critical boundary I-J singular, the variance of the moments blows up and the estimator's regime call will be unreliable. A boundary analysis or a critical-regime estimator is a natural next step.","Editorial inference: The method assumes complete ballot configurations. In many real settings only the margins of each group are available; adapting the plug-in estimator to margin-only or subsampled data is a direct extension suggested by the present construction.","Editorial inference: Proposition 8's low-temperature uniqueness is assumed from earlier articles; a reader can test it numerically for random positive-definite J without waiting for a proof."],"forward_implications":["In the high-temperature regime, the coupling matrix J is consistently estimated with error of order 1/sqrt(n), and the bias caused by finite group sizes N1,N2 disappears as the groups grow.","The estimator is computed in O(n) time from the ballot sample; there is no partition-function evaluation, so it scales to large populations.","In the low-temperature regime the sign of the cross-group coupling J12 is recovered from the sign of the off-diagonal estimate, since sign(m2)=sign(J12).","Samples can be used to declare which regime is likely: if I - N T^{-1} N fails to be positive definite, the data are classified as low-temperature and the m-estimator is used instead.","The simulations indicate that for groups of 100 voters, sample sizes near 500 lead to correct regime classification almost always, and estimates within 0.1 of the true parameters about 95% of the time."],"fun_headline_variants":["Vote margins expose two-group Curie-Weiss couplings","Ballot data decode intergroup influence parameters","Recovering group coupling from vote margin samples","Two-group Curie-Weiss reconstruction via vote margins","Estimating social cohesion from voting margin products"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is Proposition 8, stated with proof deferred to earlier articles: in the low-temperature regime there is a unique point (m1,m2) with m1>0 and m2 != 0 such that the normalized margin covariances converge to m_lambda m_nu; if that uniqueness or the sign-flip mixture structure fails for some admissible coupling matrix J, the low-temperature part of Theorem 11 collapses.","fun_headline_variants_meta":{"raw":{"variants":["Vote margins expose two-group Curie-Weiss couplings","Ballot data decode intergroup influence parameters","Recovering group coupling from vote margin samples","Two-group Curie-Weiss reconstruction via vote margins","Estimating social cohesion from voting margin products"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00022,"raw_usage":{"total_tokens":1303,"prompt_tokens":783,"completion_tokens":520,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":527,"completion_tokens_details":{"reasoning_tokens":462}},"tokens_in":527,"tokens_out":520,"duration_ms":5325,"temperature":1.0,"reasoning_tokens":462,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T09:19:52.944164+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Fix a low-temperature, positive-definite J with negative cross-coupling (for example J11=2.0, J22=1.8, J12=-1.0), simulate the two-group Curie-Weiss model for large N1=N2=N, and compute E[S1S2]/(N1N2). If the limit does not approach m1m2 with sign(m2)=sign(J12), or if two distinct (m1,m2) points both reproduce the observed product limits, Proposition 8 is false and the theorem's low-temperature branch fails.","supporting_citations":[],"review_version":1}