{"id":"e88bb3c6-9d25-4b0a-9752-03179f90d7bf","arxiv_id":"2511.19323","paper_version":2,"verdict":"REJECT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper proves loose bounds 0.288/n! 2^{(n-1)^2} < B_n < 120/n! 2^{n^2-n}, but the abstract's exact asymptotic B_n = 2^{n^2-n+1}/n!(1+o(1)) is not derived.","lead":"The paper claims to determine the asymptotic number of minimal balanced collections, but the proof body only establishes upper and lower bounds whose ratio is exponential in n. Readers should treat the headline formula in the abstract as unsupported, while the weaker bound result may still be useful in combinatorial game theory.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The abstract's asymptotic formula is not derived in the body: Theorem 1's bounds differ by 2^{Θ(n)}, and Section 7 lists closing this gap as future work.","rationale":"The reader's strongest_claim correctly identifies the abstract-versus-Theorem-1 gap, and I agree the paper should not be accepted as containing the announced asymptotic. However, the reader's weakest_assumption focuses on Lemma 6.1's F2-to-Q transfer and Lemma 5.4's mixed-sign extension. On inspection, Proposition 6.2 is true (Q-dependence would reduce mod 2 to an F2-dependence), the construction in Lemma 6.1 appears valid, and Lemma 5.4's use of the Sperner bound can be repaired via Corollary 5.1.3 (a positive matrix with the same |U(λ)| can be obtained by negating the negative coordinates). So those are not the main obstacle. The load-bearing issue is simpler and more direct: the proof never supplies a lower bound of the same exponential order as the claimed asymptotic. Even the strongest lower bound, Corollary 6.3, is smaller than the claimed leading term by a factor ~2^{-n}. The paper's own Section 7 concedes that closing this gap is future work. Credit is due for the exact small-m formulas, the small-n table, and the nontrivial bounds in Theorem 1, but those do not support the abstract. Therefore the verdict remains REJECT (UNCHANGED), while the weakest-assumption emphasis would shift to the missing matching lower bound.","tokens_in":13972,"tokens_out":16514,"duration_ms":143962,"concrete_test":"Re-derive the lower side of the claimed asymptotic from the lemmas. The strongest stated lower bound is Corollary 6.3: B_{n,n} ≥ α/n! · 2^{(n-1)^2} = α·2^{1-n}/n! · 2^{n^2-n}. Since α·2^{1-n} → 0, this cannot yield B_n ≥ (1/n!)2^{n^2-n+1}(1-o(1)). Check whether any other statement in §4 or §6 provides a lower bound with exponent n^2−n; if none exists, the abstract's (1+o(1)) formula is unproven and the paper should be reframed around Theorem 1.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim, B_n = 2^{n^2-n+1}/n!(1+o(1)), is not a consequence of anything proved in the paper. Theorem 1 gives 0.288/n! · 2^{(n-1)^2} < B_n < 120/n! · 2^{n^2-n}. Since 2^{(n-1)^2} = 2^{1-n}·2^{n^2-n}, the lower bound trails the upper bound's exponent by a factor 2^{1-n}; the upper/lower ratio is ~(120/(0.288·2^{1-n})) = O(2^n), so no (1+o(1)) asymptotic follows. This is not a subtle technical gap: Section 7 explicitly lists 'Reducing the gap between the upper and lower bounds, ideally up to a constant factor or even to (1+o(1))' as future work, directly contradicting the abstract. Corollary 6.4 gives an upper bound matching the abstract's leading term, and Lemma 6.5 shows B_n < 30B_{n,n}, so the upper side is plausible; the missing piece is a lower bound with exponent n^2−n. The body's legitimate weaker result—bounds within an exponential factor—does not establish the announced enumeration. The reader's rejection of the central claim is justified; I do not see a fatal flaw in Theorem 1 itself.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript develops a matrix-theoretic framework for counting minimal balanced collections of subsets of [n]. It proves that a collection is minimal balanced iff its 0-1 matrix has full column rank and a unique all-positive weight vector solving Mλ=1 (Lemma 3.1), derives a counting formula in terms of the set of ``unificators'' of the weight vector (Lemma 4.2), and then uses a Z2 action that inverts chosen columns to relate positive and mixed-sign full-rank matrices (Lemmas 5.1--5.4). The main proved result is Theorem 1: 0.288/n! · 2^{(n−1)^2} < B_n < 120/n! · 2^{n^2−n}. The arXiv metadata abstract, however, claims the sharper asymptotic B_n = 2^{n^2−n+1}/n! (1+o(1)). The body does not prove this; Section 7 explicitly lists reducing the exponential gap to a constant factor or to (1+o(1)) as future work. The full-text abstract on page 1 states only the bounds, so the paper also contains an internal inconsistency between the two abstracts.","tokens_in":14307,"tokens_out":13327,"duration_ms":126211,"significance":"If the asymptotic formula claimed in the arXiv abstract were established, it would be a substantial answer to a natural enumeration problem originating in cooperative game theory. The paper's structural lemmas, especially the orbit-ratio control in Lemma 5.4 and the finite-field construction in Lemma 6.1, are plausible and potentially reusable, and the exact formulas for fixed m (Corollary 4.1 and the table after it) are useful. However, the headline asymptotic claim is not a consequence of any theorem in the paper: the proved lower and upper bounds are separated by a factor exponential in n. The paper's genuine contribution at present is a nontrivial pair of bounds within an exponential factor, not the announced enumeration.","major_comments":[{"comment":"The abstract claims B_n = 2^{n^2−n+1}/n! (1+o(1)) and says the asymptotic number is determined. Theorem 1 proves only 0.288/n!·2^{(n−1)^2} < B_n < 120/n!·2^{n^2−n}. Since 2^{(n−1)^2}=2^{n^2−2n+1}, the upper/lower ratio is (120/0.288)·2^{n−1} = O(2^n), so no (1+o(1)) formula follows. Section 7, item 1 explicitly lists ``reducing the gap between the upper and lower bounds, ideally up to a constant factor or even to (1+o(1))'' as future work. Thus the paper's central announced result is not derived anywhere in the body. The full-text abstract on page 1 states only the bounds, so the arXiv metadata abstract and the article abstract also disagree; whichever is intended, the asymptotic claim is unsupported.","section":"Abstract / Theorem 1 / §7"},{"comment":"The missing ingredient for the claimed asymptotic is a lower bound with exponent n^2−n. Corollary 6.3 gives B_{n,n} ≥ α/n!·2^{(n−1)^2}, and Lemma 6.5 shows B_n < 30 B_{n,n}; combining this with Corollary 6.4 yields Theorem 1's upper bound. But no argument converts the lower exponent 2^{n^2−2n+1} into the abstract's 2^{n^2−n+1}. This is not a matter of sharpening constants; the gap between the two exponents is exponential in n. Since the authors themselves defer this to future work, the title and abstract overstate what is proved.","section":"§6, Cor. 6.3 and Lemma 6.5"}],"minor_comments":[{"comment":"The proof of Proposition 6.2 is given as ``Clear!'' but the statement is load-bearing for the lower bound. It is true for 0-1 vectors: F2-independence gives an m×m minor with determinant 1 mod 2, hence an odd integer determinant over Z, so the vectors are Q-independent. Please include this argument.","section":"§6, Prop. 6.2"},{"comment":"The zero-coordinate argument says a Q-dependence of 1 and {a'_i}_{i∉S} contradicts the earlier independence claim ``since S≠∅.'' To make this precise, also rule out S=[n]; that case is impossible because A'λ=1 would fail, but the text should say so, since the independence claim only applies to proper subsets.","section":"§6, Lemma 6.1, step 4"},{"comment":"Lemma 4.5's upper bound on |U(λ)| is proved for positive weight vectors. In Lemma 5.4 the same bound is used for mixed-sign λ. The reduction via Corollary 5.1.3 and the inversion z_{neg(λ)} works, but should be stated explicitly, because the printed proof of Lemma 4.5 uses positivity when it concludes v=u from (u−v)λ=0.","section":"§5, Lemma 5.4"},{"comment":"In the statement and first sentence of the proof, the symbol x_i appears before λ is introduced; it should be λ_i.","section":"§5, Lemma 5.2"},{"comment":"Table 3 appears to contain timing data in seconds rather than mathematical data. If it is not essential, remove it or move it to an appendix; as printed it is a distraction.","section":"Table 3"}],"recommendation":"reject","confidential_remarks":"The paper contains a salvageable core result — Theorem 1, a pair of bounds within an exponential factor — and the structural lemmas are mostly sound. But the version under review claims in its arXiv metadata abstract an exact (1+o(1)) asymptotic that the paper explicitly postpones to future work. That is a load-bearing mismatch that cannot be repaired by local revision; it requires new mathematics (a lower bound with exponent n^2−n). I would recommend reject, while noting that a resubmission reframed as a bounds paper, with the abstract corrected and the lower-bound construction fully spelled out, could be a reasonable modest contribution."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: read this as a bounds paper, not an enumeration paper. Theorem 1 — 0.288/n! · 2^{(n−1)^2} < B_n < 120/n! · 2^{n^2−n} — is new, looks correct, and the techniques are worth knowing. What the arXiv abstract claims — B_n = 2^{n^2−n+1}/n! (1+o(1)) — is not proved anywhere in the body. The paper's own internal abstract only claims the bounds, and Section 7 explicitly lists closing the gap to (1+o(1)) as future work. So the overclaim is in the metadata, not in the math.\n\nWhat's genuinely new: the Z2 action on characteristic vectors (Lemma 5.1) is a clean observation, showing each orbit of nonzero matrices has exactly two positive representatives. The lower-bound construction over F2 (Lemma 6.1) is clever: it builds a full-rank 0-1 matrix with a stronger independence property needed to guarantee a nonzero weight vector. Proposition 6.2 — F2 independence implies Q independence for 0-1 vectors — is true, and the proof label \"Clear!\" is terse but acceptable. The resulting lower bound is real, as is the upper bound via the orbit-size transfer and Stirling estimates.\n\nSoft spots, in proportion. Lemma 4.5's Sperner-bound proof is written for positive weight vectors; Lemma 5.4 applies it to nonzero matrices through Corollary 5.1.3. The transfer works because every orbit contains at least one positive representative, but the paper doesn't spell that out. Presentation gap, not a mathematical one. There are small typos (e.g., `x_i` for `λ_i` in Lemma 5.2) and the one-word proof of Proposition 6.2 is unusually terse. None of this threatens Theorem 1.\n\nThe real problem is the abstract mismatch. The bounds differ by an exponential factor, so no (1+o(1)) asymptotic follows. The body is honest about this; the metadata is not. If the abstract stays as is, readers will be misled. The paper should be reframed around the bounds result, which is a legitimate and useful contribution.\n\nWho gets value: researchers working on balanced collections, core nonemptiness, or hypergraph fractional matchings. The small-m exact formulas and the n≤6 tables are handy references. This is a solid weaker result, not the headline that the metadata advertises.\n\nI'd send it to peer review. The proof of the bounds deserves referee time, and the abstract issue is easily fixed. The reader's outright reject is too harsh.","headline":"The body proves new bounds on minimal balanced collections; the arXiv abstract's exact asymptotic is not supported by the proof and is explicitly deferred as future work.","tokens_in":14705,"tokens_out":10617,"would_cite":true,"duration_ms":92231,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05A16","05D05","91A12"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that the number B_n of minimal balanced collections of subsets of an n-element set lies between 0.288·2^{(n−1)^2}/n! and 120·2^{n^2−n}/n!, establishing B_n = 2^{Θ(n^2)}/n!, but the sharper asymptotic formula in the abstract","keywords":["minimal balanced collections","balanced collections","cooperative game core","0-1 matrices","weight vectors and unificators","Z2 inversion action","asymptotic enumeration","extremal antichain bound"],"falsifier":"Compute B_8 and B_9 exactly using the paper's inductive construction of Λ_m (as was done for n≤7 in Table 2) and compare n!·B_n / 2^{n^2−n+1}; if this ratio does not approach 1 or fluctuates, the abstract's asymptotic formula is false. Alternatively, search small m for a nonzero weight vector λ with |U(λ)| exceeding the middle binomial coefficient; such a vector would invalidate the upper-bound transfer in Lemma 5.4.","tokens_in":13905,"feed_emoji":"🧮","tokens_out":12333,"duration_ms":104169,"temperature":0.7,"pith_summary":"Minimal balanced collections—families of subsets whose weighted characteristic vectors hit the center of the cube—are the combinatorial objects underlying the non-emptiness of cooperative-game cores. This paper proves that their total number B_n satisfies 0.288·2^{(n−1)^2}/n! < B_n < 120·2^{n^2−n}/n!, showing B_n = 2^{Θ(n^2)}/n!. The argument transfers counts between full-rank 0-1 matrices with any nonzero weight vector and those with strictly positive weights, using a Z_2 action that complements columns. Most minimal collections are shown to have exactly n subsets. The abstract states the sharper asymptotic B_n = 2^{n^2−n+1}/n!(1+o(1)), but the body does not derive it; Theorem 1 is the paper's actual proved result.","feed_headline":"Minimal balanced collections: B_n has order 2^{n^2}/n!","feed_subtitle":"Explicit bounds bracket the true count; the sharper asymptotic formula claimed in the abstract is still unproved.","key_machinery":"The key object is the set U(λ)={u∈{0,1}^m : u·λ=1} of unificators of a weight vector—0-1 rows whose dot product with λ equals 1. For any balanced matrix, its rows are unificators, and λ is realizable exactly when U(λ) spans R^m (Lemma 4.1). The maximum size of U(λ) over positive λ is the middle binomial coefficient, by the extremal antichain bound (Lemma 4.5). The Z_2^m inversion action on columns changes signs of weights without changing |U(λ)|, so an orbit of a nonzero matrix has size 2^{m−|U(λ)|}, and each orbit contains exactly two positive matrices. This orbit-level bookkeeping, combined with an F_2-based construction of many nonzero matrices (Lemma 6.1) and a crude upper bound on all f","core_discovery":"The central discovery is that the number of minimal balanced collections is governed by the number of full-rank n×n 0-1 matrices whose unique weight vector has no zero coordinates, up to a factor of about 2^n and n!. Concretely, Lemma 3.1 identifies minimal balanced collections with full-rank 0-1 matrices M satisfying Mλ=1 for a strictly positive λ; then the Z_2^m action that replaces any column by its complement flips the sign of the corresponding weight, and every orbit contains exactly two positive matrices (Lemma 5.3). This yields a transfer inequality (Lemma 5.4) relating positive and nonzero matrices. A linear-algebra construction over F_2 produces enough nonzero n×n matrices for the l","pith_inferences":["The abstract's asymptotic formula B_n = 2^{n^2−n+1}/n!(1+o(1)) would require closing the gap between the lower bound ~2^{n^2−2n} and the upper bound ~2^{n^2−n}; this is a natural target for a sharper analysis of U(λ) for mixed-sign weight vectors.","The orbit method suggests a probabilistic interpretation: choosing a random 0-1 matrix and conditioning on full rank and nonzero weight, the number of positive representatives per orbit is exactly 2, which may connect to random hypergraph perfect matchings.","The F_2 construction generalizes in an obvious way to other finite fields F_{p^k}, as the paper itself lists; if the F_2 independence-to-Q transfer has analogues, similar bounds would follow for those fields.","The link to the colorful Carathéodory theorem indicates that geometrically, the count of minimal balanced collections is controlled by how many antipodal point configurations contain the origin in exactly two spanned simplices; this could yield a geometric proof of the upper bound."],"forward_implications":["B_n = n^{-1}2^{Θ(n^2)}: the leading exponent of the count is n^2, up to polynomial factors.","Most minimal balanced collections have exactly n subsets: Lemma 6.5 gives B_n < C B_{n,n} for a constant C, so the m=n term dominates the total.","The factor 2 per nonzero orbit means every minimal balanced collection can be paired with another obtained by complementing all its sets, and both remain balanced.","The lower bound's construction produces minimal collections from F_2-bases with an extra independence condition, giving a concrete infinite family.","For fixed m, the exact count B_{n,m} is computable from the set of weight vectors Λ_m via the inclusion-exclusion formula in Corollary 4.1; the paper lists closed forms for m≤4."],"fun_headline_variants":["Minimal balanced collections: B_n grows like 2^{n^2}/n!","Counting minimal balanced sets: asymptotic 2^{n^2}/n!","Minimal balanced collections: order 2^{n^2}/n! confirmed","New bounds on minimal balanced collection counts","Asymptotic enumeration of minimal balanced families"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The lower bound rests on the claim that a family of 0-1 vectors that is linearly independent over F_2 remains linearly independent over Q, so that a matrix built from such a family has a unique rational weight vector with no zero coordinates; if that transfer fails, the lower bound construction collapses—and the upper-bound transfer also relies on the middle-binomial bound applying to mixed-sign weight vectors via sign-flip invariance.","fun_headline_variants_meta":{"raw":{"variants":["Minimal balanced collections: B_n grows like 2^{n^2}/n!","Counting minimal balanced sets: asymptotic 2^{n^2}/n!","Minimal balanced collections: order 2^{n^2}/n! confirmed","New bounds on minimal balanced collection counts","Asymptotic enumeration of minimal balanced families"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000181,"raw_usage":{"total_tokens":1107,"prompt_tokens":673,"completion_tokens":434,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":417,"completion_tokens_details":{"reasoning_tokens":344}},"tokens_in":417,"tokens_out":434,"duration_ms":4735,"temperature":1.0,"reasoning_tokens":344,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T20:31:42.035957+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute B_8 and B_9 exactly using the paper's inductive construction of Λ_m (as was done for n≤7 in Table 2) and compare n!·B_n / 2^{n^2−n+1}; if this ratio does not approach 1 or fluctuates, the abstract's asymptotic formula is false. Alternatively, search small m for a nonzero weight vector λ with |U(λ)| exceeding the middle binomial coefficient; such a vector would invalidate the upper-bound transfer in Lemma 5.4.","supporting_citations":[],"review_version":1}