{"id":"72b9dccd-26ed-4f05-9663-5c08a3cc46a2","arxiv_id":"2505.21950","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"For every n≥3, every rational p between 1/2 and the bound π_n is the common win probability of some balanced nontransitive n-tuple of dice.","lead":"A mathematics paper proves that for any number of dice n from 3 upward, every rational winning chance up to a known limit can be produced by a circular set of dice where each die usually beats the next. The result completes the answer to a long-standing puzzle about nontransitive dice.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 3.2's transposition schedule is not formalized; the claim that every intermediate count in the interval is attained needs a precise algorithm or verification.","rationale":"The reader's weakest_assumption coincides with mine: the proof of Lemma 3.2 is the hinge of Theorem 1.2. All the algebra in Section 3 reduces the theorem to applying Lemma 3.2 with s_j=pm^2, so the lemma must be fully reliable. The inductive proof in Section 4.1 is plausible—the two phases (A_j's before the A_{j+1} block crossing A_{j-1}'s in τ1, and A_j's after the block crossing A_{j-1}'s in τ2) do add a_j·a_{j−1} and (m_j−a_j)(m_{j−1}−a_{j−1})—but the text does not define the scheduling precisely enough to prove that no integer is skipped. This is an internal rigor gap, not a demonstrated counterexample; my proposed brute-force check would distinguish 'lemma false' from 'proof needs expansion.' I also noted a minor indexing issue: Lemma 3.3 and Lemmas 4.3/4.4 index auxiliary variables p_j starting at j=2, so for k=1 (n=3 or 4) the statements reference an undefined p_1; this is a typo since the n=4 construction can be checked directly and n=3 is already known, and it does not change the verdict. The paper's main claim is credible and the upper bound from [2,9] is used correctly; the verdict should remain conditional pending a rigorous write-up of the transposition argument.","tokens_in":13072,"tokens_out":31859,"duration_ms":299001,"concrete_test":"Brute-force check: for n=4 and n=5 with m_i=3 or 4, enumerate all central words of each type (a_1,...,a_{n-1}) and compute the attainable set of (N(A_1<A_2),...,N(A_{n-2}<A_{n-1})). Verify that this set is exactly the box (2). If any integer point in the box is missing, Lemma 3.2 is false; if all appear, the concern is about proof rigor only, and a written scheduling invariant would settle it.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The main construction in Theorem 1.2 relies on Lemma 3.2, which guarantees a central (m_1,...,m_n)-word of prescribed type realizing every vector (s_1,...,s_{n-2}) in the box (2). In the inductive step of Section 4.1, starting from σ0 = A_j^{a_j} τ1 A_{j+1}^{m_{j+1}} A_j^{m_j-a_j} τ2, the proof defines σ1,...,σγ by performing 'as many transpositions A_jA_α→A_αA_j for α≤j−2 as possible', then 'one transposition A_jA_{j−1}→A_{j−1}A_j'. It asserts that each iteration increases q_{j−1}=N(A_{j−1}<A_j) by exactly 1 and preserves q_1,...,q_{j−2}, but it never specifies which A_j is moved, what 'as possible' means, or why, after a maximal block of A_α swaps, an adjacent A_jA_{j−1} pair is always available. If 'as many as possible' is read globally, an A_j can be moved past all A_α's without encountering an A_{j−1} before the central block, making the next step impossible. The conclusion that q_{j−1} takes every integer value in [a_{j−1}(m_j−a_j), a_{j−1}m_j+(m_j−a_j)(m_{j−1}−a_{j−1})] is therefore an unproven scheduling claim. Since Theorem 1.2 needs s_j=pm^2 for an arbitrary rational p, a single skipped integer would break the characterization.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies balanced and nontransitive n-tuples of dice (BN n-tuples) and characterizes the set of achievable common winning probabilities. For n≥3 and every rational p in (1/2, π_n], where π_n = 1 − 1/(4 cos^2(π/(n+2))), the paper constructs a BN n-tuple of dice, each with the same number of sides, whose winning probability equals p. Together with the known upper bound of Bogdanov and Komisarski, this gives a complete classification of the achievable winning probabilities for BN n-tuples of dice. The proof represents dice by words, reduces the construction to a word-combinatorial lemma (Lemma 3.2), and uses a rational approximation argument based on trigonometric quantities from Komisarski's work (Lemma 3.3).","tokens_in":1645,"tokens_out":1871,"duration_ms":387800,"significance":"If the proof is made fully rigorous, this is a complete resolution of Problem 1.4 posed by Kim et al., extending the known n=3 classification to all n≥4 and also covering n=3. The paper's construction is explicit and combinatorial, and the endpoint case p=π_n is handled with concrete computations. The re-derivation of the needed properties of Komisarski's auxiliary values in Lemmas 4.3 and 4.4 is elementary and checkable. The main value added is the matching lower-bound construction, since the upper bound is external. The paper also ships the key reduction in a clean form, which should make the result usable for further questions such as Problem 5.1.","major_comments":[{"comment":"The proof of Lemma 3.2 is not fully specified. In the inductive step, σ_{i+1} is defined by 'Perform as many transpositions A_jA_α → A_αA_j for α ∈ {1,...,j−2} as possible' followed by 'Perform one transposition A_jA_{j−1} → A_{j−1}A_j'. This does not identify which copy of A_j is moved, whether the swaps are applied in the left or right region relative to the A_{j+1} block, or in what order. The assertion that after step 2 an adjacent A_jA_{j−1} pair is available is not proved; a global reading of 'as many as possible' could move A_j's past A_α's in one region without producing an A_jA_{j−1} adjacency where it is needed. Consequently, the claim that q_{j−1} takes every integer value in the stated interval is an unproven scheduling statement. Since Lemma 3.2 is the engine of Theorem 1.2, this must be repaired, for instance by giving an explicit algorithm that moves a chosen A_j rightward past A_α's until it reaches an A_{j−1}, performs the swap, and then proceeds, in each region, and by proving that this exhausts the interval.","section":"Section 4.1"},{"comment":"Lemma 3.3 and its use in the proof of Theorem 1.2 have an indexing gap for n=3 and n=4. For these n we have k=floor((n−1)/2)=1, so the tuple p_2,...,p_k is empty, yet the last line of the inequalities (3) uses p_1, and the proof of Theorem 1.2 goes on to define a_ℓ using p_{n−(k+1)} etc., which is only meaningful for k≥2. Thus the proof as written does not cover the case n=4 with p<π_4, even though n=4 is part of the theorem and of the motivating problem. This is fixable by treating n=3 via Theorem 1.3 and n=4 via the same direct construction used for p=π_4 (that construction works for all p∈(1/2,2/3]), or by extending the definition of p_j to j=1. As written, however, the proof is incomplete for these cases.","section":"Sections 3 and 4.2"}],"minor_comments":[{"comment":"The typeset text contains many garbled fractions and broken expressions, such as 'greater than1\\n2', 'πn := 1 − 1\\n4 cos2( π\\nn+2 )', and 'p ∈\\n 1\\n2 , πn\\n'. This makes the paper unnecessarily hard to read and should be corrected in the final version.","section":"Throughout"},{"comment":"The examples in Observation 4.2 appear garbled. For Case 1, swapping A_3A_1 to A_1A_3 in σ=A3 A1 A3 A4 A3 A2 A3 gives A1 A3^2 A4 A3 A2 A3, not the word printed. Similar corruption appears in the Case 2 and Case 3 examples.","section":"Section 4.1"},{"comment":"The sentence 'Since cos(α) < 1 < 1/2−√2' is unintelligible as printed. The inequality being proved for the odd-n case ((1−p*_k)^2 < 1/2) is true, but the printed argument is not understandable and should be rewritten.","section":"Lemma 4.3"},{"comment":"The notation 'there exists rationals p_2,...,p_k' is problematic when k=1, since the list is empty but the displayed inequalities still refer to p_1 in the last line. The statement should either require n≥5 or define p_1 for the small cases.","section":"Lemma 3.3"},{"comment":"There are small typos: 'nontransitve' in the abstract, and 'P(A1, < A2)' with an erroneous comma in the Introduction.","section":"Abstract and Introduction"}],"recommendation":"major_revision","confidential_remarks":"The core construction appears sound and the gaps are fixable within the scope of the manuscript: Lemma 3.2 needs a precise transposition schedule or a more formal connectivity argument, and the n=3,4 indexing in Lemma 3.3 needs a small-case treatment. I therefore recommend major revision rather than rejection. I would ask the editor to ensure the revised version either proves the scheduling claim in Lemma 3.2 or replaces it with a fully rigorous argument, and to check that the theorem's proof covers n=4 for all rational p in (1/2, 2/3)."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nYou should know two things about this paper quickly. First, it solves Kim et al.'s problem on balanced nontransitive dice: for every n≥3 and every rational p in (1/2, π_n], there is a balanced nontransitive n-tuple of dice with winning probability p. Together with the Bogdanov–Komisarski upper bound, this is the full characterization. Second, the proof is mostly checkable, but the key lemma, Lemma 3.2, has an under-specified transposition step that needs a careful rewrite before I would call it rigorous.\n\nThe new contribution is genuine. The n=3 case was already classified; here the construction extends to all n, and the paper handles the endpoint p=π_n (which forces n=4) as a separate case. Lemma 3.2 generalizes Kim et al.'s pairwise-count lemma to n dice, and Lemma 3.3 uses Komisarski's p_j^* values plus a density argument to find rational auxiliary parameters. The algebra for the construction lines up cleanly, and the paper does not overclaim: it states the rationals in the interval as the achievable set, consistent with the upper bound.\n\nThe soft spot is in Section 4.1. The inductive step says to apply 'as many transpositions A_j A_α -> A_α A_j as possible' and then one A_j A_{j-1} -> A_{j-1} A_j, repeating until the count q_{j-1} reaches the target. That is a bubble-sort idea, but the direction is not specified: which A_j moves, what 'as possible' means, and why an A_{j-1} is always adjacent after the A_α moves. If the reader interprets 'as possible' globally, an A_j could shoot past all the small-index letters and never meet an A_{j-1}. I think the lemma is true and the interval formula is consistent with the invariants, so this is a rigor gap in the write-up, not a fundamental problem. The typesetting is also rough: several fractions in the displayed equations are garbled in the arXiv text, which made checking Lemmas 4.3 and 4.4 slower. Probably a conversion artifact, but it should be fixed.\n\nThis paper is for anyone working on nontransitive dice or combinatorial probability, and it is a nice application of a density argument to turn an upper bound into a constructive classification. It deserves a serious referee. My recommendation: send it to review, ask the author to formalize the transposition schedule in Lemma 3.2 and clean up the rendering, then accept.","headline":"Settles the Kim et al. open problem with a full classification of achievable winning probabilities; proof is sound overall, but Lemma 3.2's transposition argument needs a careful rewrite.","tokens_in":13963,"tokens_out":9013,"would_cite":true,"duration_ms":84478,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60C05"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that every rational number in (1/2, π_n] can be realized as the common winning probability of a balanced nontransitive n-tuple of dice, completing the classification.","keywords":["nontransitive dice","balanced dice","winning probability","central words","adjacent transpositions","least upper bound","rational density","combinatorial probability"],"falsifier":"One could enumerate, for a small pair $(n, m)$ and a fixed type $(a_1,\\dots,a_{n-1})$, all central words and compare the attainable values of $N_\\sigma(A_{j-1} < A_j)$ with the interval claimed in Lemma 3.2; a single missing integer would falsify the lemma. A more targeted test runs the inductive transposition process from Section 4.1 on a concrete example and checks whether the count increases by exactly one at each step or jumps over any value.","tokens_in":12860,"feed_emoji":"🎲","tokens_out":8920,"duration_ms":74507,"temperature":0.7,"pith_summary":"Balanced and nontransitive dice are arranged in a cycle so that each die beats its successor with probability greater than $1/2$, and all these probabilities are equal. This paper proves that for every integer $n \\geq 3$ and every rational $p$ strictly between $1/2$ and the bound $\\pi_n = 1 - \\frac{1}{4\\cos^2(\\pi/(n+2))}$, there exists a balanced nontransitive $n$-tuple of dice whose common winning probability is exactly $p$. Combined with the known upper bound $w \\leq \\pi_n$ [2, 9], this fully characterizes the set of achievable winning probabilities and answers the question posed in [7]. The result matters because it turns a long-studied probabilistic paradox into a precisely understood phenomenon: the winning chance can be tuned to any rational value up to a sharp ceiling.","feed_headline":"Balanced nontransitive dice realize every rational winning chance","feed_subtitle":"The winning probabilities of balanced nontransitive n-tuples are now fully classified, closing a problem left open in 2024.","key_machinery":"The carrying object is the central word: a word $\\sigma = \\sigma(A_1,\\dots,A_n)$ in which all faces of $A_n$ appear in a single contiguous block, with type $(a_1,\\dots,a_{n-1})$ recording how many of the letters $A_1,\\dots,A_{n-1}$ lie before that block. Lemma 3.2 extends a lemma of [7] to $n \\geq 4$: for any prescribed counts $s_i$ in the natural intervals, a central word of a given type exists with $N_\\sigma(A_i < A_{i+1}) = s_i$ for each $i = 1,\\dots,n-2$. The proof proceeds by induction, inserting the new die as a block and then applying adjacent transpositions that change exactly one comparison count at a time. Lemma 3.3 uses the continuity of the bracketing functions $L$ and $U$ and the density of the rationals to produce rational parameters $p_2,\\dots,p_k$ that keep the prescribed target $p$ inside the required interval, so that the integer-interval conditions of Lemma 3.2 hold for a suitably large even $m$.","core_discovery":"The paper's central claim is Theorem 1.2: for every $n \\geq 3$ and every rational $p \\in (1/2, \\pi_n]$, there is a positive integer $m$ and a balanced nontransitive $n$-tuple of $m$-sided dice $(A_1,\\dots,A_n)$ with $w(A_1,\\dots,A_n) = P(A_1 < A_2) = \\cdots = P(A_n < A_1) = p$. The proof treats the endpoint separately, showing that $\\pi_n$ is rational only when $n = 4$, where $\\pi_4 = 2/3$ and the construction follows directly from Lemma 3.2. For $p < \\pi_n$, the proof selects rational parameters $p_2,\\dots,p_k$ by a continuity and density argument around the trigonometric values defined in [9], then applies the extended word lemma to realize the desired comparison counts.","pith_inferences":["The informal 'step 2' in the proof of Lemma 3.2 deserves a rigorous justification; if a gap exists, a different rearrangement scheme might still realize every integer count, so the main theorem need not fall.","The trigonometric parameters used in Lemma 3.3 are ratios of consecutive sine values, hinting at a spectral interpretation of the bound $\\pi_n$ that could extend to other cyclic comparison problems.","The construction gives no bound on the minimal number of sides $m$ needed for a given rational $p$; numerical experiments near $p = \\pi_n$ could reveal how that minimal $m$ grows."],"forward_implications":["The achievable winning probabilities of balanced nontransitive $n$-tuples are exactly the rationals in $(1/2, \\pi_n]$, so the set of attainable values is dense in that interval.","Problem 1.4 of [7] is settled: the least upper bound for $n \\geq 4$ is $\\pi_n$, and it is attained only for $n = 4$, where $\\pi_4 = 2/3$.","Every rational winning probability in the range can be realized using dice that all have the same number of sides $m$, so unequal side counts are never necessary for exact realizations.","Since $\\pi_n < 3/4$ for every $n$, no balanced nontransitive collection of dice can have a common winning probability of $3/4$ or more, no matter how many dice are used.","The same word-tuning machinery is pointed out by the author as a likely route toward the open side-count problem posed in [7]."],"supporting_citations":[{"why":"Provides Lemma 3.1 that Lemma 3.2 extends, the triple classification, and the open problem (Problem 1.4) that motivates the paper.","marker":"[7]"},{"why":"Establishes the upper bound $w \\leq \\pi_n$ and defines the trigonometric $p^*_j$ values around which the rationality argument is built.","marker":"[9]"},{"why":"Independently proves the same closed-form upper bound $\\pi_n$ for the winning probability.","marker":"[2]"},{"why":"Gives the $n = 3$ upper bound $(\\sqrt{5}-1)/2$, the base case of the bound studied here.","marker":"[16]"},{"why":"Provides the earlier implicit upper bound for $n$-tuples of random variables that the explicit bound refines.","marker":"[17]"},{"why":"Introduces the study of balanced nontransitive dice and asks whether the triple bound can be improved.","marker":"[14]"}],"fun_headline_variants":["Balanced nontransitive dice hit every rational win chance","All rational winning probabilities realized for n≥3 dice","Dice puzzle solved: rational odds up to π_n are possible","Complete win-probability range for balanced nontransitive dice"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the induction step of Lemma 3.2, which assumes that adjacent transpositions can move the comparison count $N_\\sigma(A_{j-1} < A_j)$ through every integer in its claimed interval while preserving all earlier counts; the informal description of that step does not rigorously rule out skipped values.","fun_headline_variants_meta":{"raw":{"variants":["Balanced nontransitive dice hit every rational win chance","All rational winning probabilities realized for n≥3 dice","Dice puzzle solved: rational odds up to π_n are possible","Complete win-probability range for balanced nontransitive dice"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000366,"raw_usage":{"total_tokens":2072,"prompt_tokens":1155,"completion_tokens":917,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":771,"completion_tokens_details":{"reasoning_tokens":848}},"tokens_in":771,"tokens_out":917,"duration_ms":8098,"temperature":1.0,"reasoning_tokens":848,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T13:30:09.189487+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"One could enumerate, for a small pair $(n, m)$ and a fixed type $(a_1,\\dots,a_{n-1})$, all central words and compare the attainable values of $N_\\sigma(A_{j-1} < A_j)$ with the interval claimed in Lemma 3.2; a single missing integer would falsify the lemma. A more targeted test runs the inductive transposition process from Section 4.1 on a concrete example and checks whether the count increases by exactly one at each step or jumps over any value.","supporting_citations":[{"cited_title":"Balanced nontransitive dice: Existence and probability.The Electronic Journal of Combinatorics, 31:Not available, 1 2024","cited_arxiv_id":null,"evidence_quote":"Provides Lemma 3.1 that Lemma 3.2 extends, the triple classification, and the open problem (Problem 1.4) that motivates the paper."},{"cited_title":"Nontransitive random variables and nontransitive dice.The American Mathematical Monthly, 128(5):423–434, 2021","cited_arxiv_id":null,"evidence_quote":"Establishes the upper bound $w \\leq \\pi_n$ and defines the trigonometric $p^*_j$ values around which the rationality argument is built."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Independently proves the same closed-form upper bound $\\pi_n$ for the winning probability."},{"cited_title":"Trybu la","cited_arxiv_id":null,"evidence_quote":"Gives the $n = 3$ upper bound $(\\sqrt{5}-1)/2$, the base case of the bound studied here."},{"cited_title":"Trybu la","cited_arxiv_id":null,"evidence_quote":"Provides the earlier implicit upper bound for $n$-tuples of random variables that the explicit bound refines."},{"cited_title":"Balanced nontransitive dice.The College Mathematics Journal, 48:10–16, 1 2017","cited_arxiv_id":null,"evidence_quote":"Introduces the study of balanced nontransitive dice and asks whether the triple bound can be improved."}],"review_version":1}