{"id":"36932530-827d-4d95-8ec7-9ae3862e3c95","arxiv_id":"2607.08712","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":3,"one_line_summary":"The Cramér random model for prime gaps satisfies Gilbreath's conjecture almost surely, and the only deterministic obstructions are long zero-blocks or very long shallow two-valued blocks.","lead":"This paper proves that a probabilistic model of prime gaps satisfies Gilbreath's conjecture, and gives a deterministic inverse theorem identifying the exact structural obstructions that would prevent the conjecture from holding for actual primes. It matters because it clarifies what would need to be ruled out to resolve a 66-year-old open problem.","discovery_kind":"unclear","skeptic_critique":{"model":"glm-5.2","headline":"No significant objection identified. The probabilistic core (Theorem 1.3) is self-contained; the deterministic theorem's reliance on unproved hypotheses is openly acknowledged and does not affect the rigor of the random-model results.","rationale":"The reader's verdict of ACCEPT at HIGH confidence is appropriate. The paper delivers three distinct, self-contained results: (1) Theorem 1.2 proving the Cramér random model satisfies Gilbreath's conjecture a.s., (2) the general Theorem 1.3 with the sharp 2-separated-set obstruction, and (3) the deterministic inverse theorem (Theorem 1.6) identifying structural obstructions. The probabilistic proofs are rigorous: the tower framework, the separation lemma, the large-shadow lemma, and the combinatorial counting all fit together correctly, and the Borel-Cantelli step is standard. The deterministic theorem is honestly presented as conditional on hypotheses the authors themselves describe as difficult. The continuous-model lower bound (Theorem 1.4) is a clean application of Jensen's inequality. The reader correctly identifies that the deterministic theorem's applicability to actual primes requires unproved assumptions, but this does not undermine the correctness of what is actually proven. No internal inconsistency, circularity, or hidden parameter fitting was found. The gap between the lower bound δn and the upper bound 2^{n+1} in Theorem 1.3's growth condition is acknowledged as open and related to the poorly understood sequence c_i. This is honest, substantive progress.","tokens_in":29180,"tokens_out":807,"duration_ms":764520,"concrete_test":"Independently verify the tower-counting bound in Lemma 4.4 by enumerating all valid towers for small parameters (e.g., n=8, D=4) and confirming that the formula (D-1 choose k)(n-1 choose k-1)(n+k-1 choose k-1) provides a valid upper bound for each k. This would catch any combinatorial miscount in the stars-and-bars argument.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central rigorous claim is Theorem 1.3 (and its corollary Theorem 1.2). The proof proceeds by defining a tower structure (Definition 3.12), showing via Lemma 3.13 that any failure event a_{(n-1,1)} > 1 produces an attained tower, bounding the probability of each tower's attainment using the Separation Lemma (Lemma 3.11) and the large-shadow property (Lemma 3.14), and bounding the number of towers combinatorially (Lemma 4.4). I checked the logical chain: the union bound in the proof of Proposition 4.1 combines these correctly, and the Borel-Cantelli application in the proof of Theorem 1.3 is standard. The key step where a_{i_s} is conditioned on a_{i_s+1},...,a_n and shown to lie in a 2-separated set (invoking Lemma 3.11) is valid because {0,d} with d>=2 is 2-separated. The combinatorial bound in Lemma 4.4 is loose (discards the non-increasing right-position constraint) but sufficient. The deterministic Theorem 1.6 is explicitly conditional on unproved axioms (i)-(iii), which the authors state clearly. The continuous model lower bound (Theorem 1.4) follows cleanly from Jensen's inequality applied to the conditional expectation. I do not find an internal inconsistency or a gap in the probabilistic argument. The reader's weakest_assumption correctly identifies that the deterministic theorem's applicability to actual primes depends on unproved hypotheses, but this is a limitation of scope, not a defect in the proofs that are actually given.","agreement_with_reader":"agree"},"referee_report":{"model":"glm-5.2","summary":"This paper studies Gilbreath's conjecture through three lenses: (1) a Cramér random model (Theorem 1.2), in which normalized prime gaps are replaced by independent geometric random variables of logarithmic size, and the left diagonal of the resulting Gilbreath array is shown to be {0,1}-valued almost surely after finitely many rows; (2) a general random-model theorem (Theorem 1.3) subsuming Theorem 1.2 and a prior result of the first author, requiring only that the random variables grow at most linearly and do not concentrate in 2-separated sets; (3) a continuous exponential model (Section 2, Theorem 1.4) for which a lower bound on the expected row-values c_i is proved via Jensen's inequality; and (4) a deterministic inverse theorem (Theorem 1.6) showing that, assuming a Cramér-type bound on initial values, the only obstructions to Gilbreath's conjecture are long zero-blocks or long shallow {0,d}-valued blocks. The probabilistic proofs proceed via a tower construction (Definition 3.12), a separation lemma (Lemma 3.11), a large-shadow lemma (Lemma 3.14), and a union bound with Borel–Cantelli. The deterministic argument uses coarse monotonicity (Lemma 5.2), good blocks (Definition 5.4), and a pigeonhole-based dichotomy (Lemma 5.8).","tokens_in":29640,"tokens_out":1813,"duration_ms":348366,"significance":"Theorem 1.2 provides the first rigorous verification of Gilbreath's conjecture for a random model with the correct logarithmic scale for prime gaps, improving on the first author's prior work [1] where the random variables grew too slowly. The general Theorem 1.3 is a clean, self-contained result with a well-motivated axiom (non-concentration in 2-separated sets) that precisely identifies the key obstruction. The deterministic inverse theorem (Theorem 1.6), while conditional on unproved hypotheses, makes the heuristic obstructions rigorous and could guide future work on the actual conjecture. The continuous model analysis and the lower bound of Theorem 1.4 (showing c_i cannot decay faster than 1/i) is a modest but interesting contribution that connects to the optimality of the linear growth threshold in Theorem 1.3. The paper is largely self-contained, with the probabilistic core verifiable from first principles.","major_comments":[{"comment":"In the proof of Proposition 4.1 (specifically the derivation following Lemma 4.4), the bound on the sum over k uses the inequality (n-1 choose k-1)(n+k-1 choose k-1) <= (n+D choose D)^2, and then the binomial theorem gives a factor of 2^D. However, the final stated bound in Proposition 4.1 is 2^D * (en/D + e)^{2D} * (prod rho_i)^{1/2}. The intermediate bound gives 2^D * (n+D choose D)^2, and using (n+D choose D) <= (e(n+D)/D)^D would yield 2^D * (e(n+D)/D)^{2D}. The appearance of the '+e' term inside the parentheses (i.e., (en/D + e)^{2D} rather than (e(n+D)/D)^{2D}) should be clarified — it may arise from a slightly different estimate, but as written the reader cannot easily trace the exact algebraic step. This is a presentation issue in a load-bearing estimate, not an error, but it should be made explicit.","section":null},{"comment":"Theorem 1.6, axiom (iii): the condition requires 2^{M-m} < d <= 2^{M-m+1}, but the proof in Section 5 produces d* in {2^{M-m}+1, ..., 2^{M-m+1}} (Proposition 5.3) and good blocks with non-zero value d in {d*, ..., 2^{M-m+1}} (Definition 5.4). The final conclusion (¬iii) states 2^{M-m} < d <= 2^{M-m+1}, which is consistent, but the relationship between the d in the conclusion and the d* from Proposition 5.3 should be stated more explicitly in the proof of the final step (Section 5.4), since the good block's non-zero value d' satisfies d' >= d >= d* > 2^{M-m}, and this chain is what yields the strict inequality in (¬iii).","section":null}],"minor_comments":[{"comment":"Section 1.3: the values c_0=1, c_1=1, c_2=7/9, c_3=227/288 are given. The Monte Carlo simulation in Figure 1 is mentioned but the figure caption could note the error bars or sample variance, since c_3 is very close to c_2 and the non-monotonicity claim relies on precise values.","section":null},{"comment":"Lemma 3.8 is referred to as 'Theorem 3.8' in its proof (e.g., 'From Theorem 3.7(i)') and similarly Lemma 3.7 is called 'Theorem 3.7' in several places (e.g., proof of Lemma 3.8, proof of Lemma 5.5). These should be 'Lemma' throughout.","section":null},{"comment":"Section 1.2, footnote 2: the sentence beginning 'The refined Cramér–Granville random model...' is slightly awkward; consider rephrasing for clarity.","section":null},{"comment":"Definition 3.12, axiom (ii): 'the bottom vertex I_{j,+inf} of nabla_{I_j} lies in the parent I_{j-1}^- of I_{j-1}' — the notation I_{j-1}^- was defined as the parent of I_{j-1}, but it would help to remind the reader that this means I_j is a subblock of I_{j-1}^-.","section":null},{"comment":"Remark 4.5: 'at least than 2^{n+1}' should be 'at least 2^{n+1}'.","section":null},{"comment":"The paper uses 'Theorem X.Y' and 'Lemma X.Y' interchangeably in cross-references within proofs (e.g., 'Theorem 5.2' for Lemma 5.2, 'Theorem 5.5' for Lemma 5.5, 'Theorem 5.7' for Lemma 5.7, 'Theorem 5.8' for Lemma 5.8). This is systematic and should be corrected throughout Section 5.","section":null},{"comment":"Section 5.1, Lemma 5.1: the inductive claim involves a parameter d that ranges over {1, ..., 2^M}, but the inductive step on d' > d uses the fact that d' <= 2^M (from axiom (i)). This is correct but the bound on d' should be stated explicitly in the inductive step for completeness.","section":null},{"comment":"The reference [3] (Eppstein, blog post) is cited for the claim that Cramér-type bounds alone are insufficient. This is a blog post rather than a peer-reviewed source; while acceptable for a heuristic claim, the authors might consider whether a more formal reference exists.","section":null}],"recommendation":"minor_revision","confidential_remarks":"The paper is a solid contribution. The probabilistic results (Theorems 1.2 and 1.3) are rigorous and self-contained. The deterministic Theorem 1.6 is conditional on unproved hypotheses, but this is openly acknowledged and does not affect the correctness of what is actually proved. The minor comments are mostly cross-referencing errors (Lemma vs. Theorem) and one algebraic clarification in the proof of Proposition 4.1. The bar for minor revision is appropriate: the central claims are sound, and the issues are local and presentational. The authors include Terence Tao, which may explain the high quality of the overall structure, but the systematic Lemma/Theorem cross-reference errors suggest a proofreading pass is needed."},"author_rebuttal":null,"desk_editor":{"model":"glm-5.2","letter":"The main thing to know: this paper proves that Gilbreath's conjecture holds for the Cramér random model — geometric random variables with logarithmic parameter — almost surely. That's a real step beyond Chase's earlier result [1], which needed sub-log-logarithmic growth and couldn't model actual prime gaps. The general Theorem 1.3 is the engine: independent non-negative integer-valued random variables with (i) eventual linear growth bound and (ii) non-concentration in 2-separated sets give {0,1}-valued left diagonal almost surely. The 2-separated-set condition is the right obstruction — if your data lives in the even integers, Gilbreath can fail, and this is essentially the only obstruction the theorem needs to rule out. The proof is clean. The tower construction (Lemma 3.13) extracts structure from any failure event, the large-shadow lemma (Lemma 3.14) forces many top-row entries into the tower's shadow, the separation lemma (Lemma 3.11) makes each such entry exponentially unlikely, and a union bound plus Borel–Cantelli finishes. I checked the logical chain and it holds together. The combinatorial count in Lemma 4.4 is loose but sufficient. The continuous-model lower bound (Theorem 1.4) via Jensen is a nice bonus — the non-monotonicity of the $c_i$ sequence is curious but not load-bearing. The deterministic inverse theorem (Theorem 1.6) is the soft spot, but it's an honest soft spot. It identifies long zero-blocks and long shallow {0,d}-valued blocks as the only obstructions, conditional on a Cramér-type gap bound and two structural hypotheses the authors openly state they cannot verify even assuming Hardy–Littlewood. This is a limitation of scope, not a defect in what's actually proved. The gap between the $O(n)$ lower bound and $O(2^n)$ upper bound on allowable growth in Theorem 1.3 is large, and the authors flag it. The reader's assessment is accurate across the board. The stress-test found no internal inconsistency, and I agree — the probabilistic argument is self-contained and correct. This is for probabilistic and combinatorial number theorists. It deserves a serious referee.","headline":"Solid paper. The Cramér-model Gilbreath result is genuinely new and correctly proved; the deterministic inverse theorem is conditional but conceptually valuable.","tokens_in":30003,"tokens_out":1676,"would_cite":true,"duration_ms":59579,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11A41","11K99"],"pacs":[],"model":"glm-5.2","headline":"Gilbreath's conjecture holds for Cramér random primes","keywords":["Gilbreath conjecture","Cramér random model","prime gaps","absolute differences","inverse theorem","2-separated sets","Borel-Cantelli lemma","geometric distribution"],"falsifier":"If normalized prime gaps were shown to concentrate in some 2-separated set (violating axiom (ii) of Theorem 1.3), or if long blocks of zeroes or long shallow {0,d}-valued blocks were shown to occur in the prime gap Gilbreath array, the path to proving the original conjecture via this paper's framework would be blocked.","tokens_in":29342,"feed_emoji":"🔢","tokens_out":1308,"duration_ms":145116,"temperature":0.7,"pith_summary":"Gilbreath's conjecture says that if you write the primes in a row, then repeatedly take absolute differences of adjacent entries to build a triangle, the left edge of that triangle is all ones after the first row. This paper proves the conjecture is true in the Cramér random model — the standard probabilistic model for primes where each gap is an independent geometric random variable of logarithmic size. The authors establish a more general result: any sequence of independent non-negative integer random variables that grow at most linearly and do not concentrate in any 2-separated set (a set with no two consecutive integers, like the even numbers) will almost surely produce a Gilbreath array whose left edge eventually takes only the values 0 or 1. The proof works by showing that if the left edge fails to reach {0,1}, this forces a tower of nested triangular regions whose values are trapped in 2-separated sets like {0,d}, and the probability of so many values being so constrained is exponentially small. The paper also gives a deterministic inverse theorem: assuming only a Cramér-type bound on prime gaps, the only things that could possibly prevent Gilbreath's conjecture from holding are long runs of zeroes or very long shallow blocks taking just two values {0,d} with d at least 2.","feed_headline":"Gilbreath's conjecture holds for Cramér random primes","feed_subtitle":"Probabilistic model of prime gaps satisfies the 66-year-old conjecture; deterministic analysis pinpoints the only possible obstructions for실","key_machinery":"The argument combines three ingredients: (1) a tower construction that decomposes any failure of the Gilbreath property into nested {0,d}-valued triangular regions, (2) a separation lemma showing that the set of top-row values forcing a given location to be {0,d}-valued is always 2-separated, and (3) a counting bound on the number of possible towers, which grows like n^{O(D)} where D is the maximum entry size. When D is at most δn for small δ, the exponential decay from the separation constraints dominates the polynomial growth of tower count, and Borel-Cantelli gives almost sure convergence to {0,1}.","core_discovery":"The central mechanism is a tower argument. When the bottom entry of a Gilbreath triangle exceeds 1, one can trace backward through the array and find a nested sequence of triangular regions, each constrained to take only two values 0 and d for some d. Each triangle in this tower casts a shadow onto the top row of the array, and the separation lemma shows that the values in the top row needed to make each triangle {0,d}-valued form a 2-separated set. Since the random variables are assumed not to concentrate in any 2-separated set, the probability of satisfying all these constraints simultaneously is a product of factors each bounded below 1, yielding an exponentially small failure probability","pith_inferences":["The non-concentration axiom (ii) in Theorem 1.3 is verified for the Cramér model but not for actual primes; if one could show that normalized prime gaps do not concentrate in any 2-separated set, the probabilistic part of the argument would transfer to the deterministic setting.","The lower bound on c_i summing to log n suggests that even in the continuous (exponential) model, the expected contraction of Gilbreath arrays is only logarithmic over n rows — meaning the array needs roughly n rows to shrink by a factor of n, which is consistent with but does not by itself prove the conjecture.","The irregular non-monotone behavior of the constants c_i (which first decrease then increase) may reflect discrete arithmetic structure analogous to the binary-digit-dependent behavior of Pascal's triangle modulo 2, suggesting deeper number-theoretic structure in the difference operator."],"forward_implications":["The general Theorem 1.3 applies to any random model satisfying two simple axioms — sublinear growth and non-concentration in 2-separated sets — so any future prime gap model meeting these criteria automatically satisfies the Gilbreath property.","The deterministic inverse theorem (Theorem 1.6) reduces the original Gilbreath conjecture to ruling out two specific combinatorial configurations in the prime gap array, giving concrete targets for future work on the actual primes.","The continuous model analysis reveals that the expected row values c_i satisfy a lower bound summing to log n, meaning the decay is at best 1/i — this constrains how fast Gilbreath arrays can contract and suggests the linear growth threshold in Theorem 1.3 may be near-optimal.","The gap between the proven lower bound δn and the counterexample at 2^n for the growth threshold in Theorem 1.3 highlights the mysterious behavior of the constants c_i as a key open problem."],"fun_headline_variants":["Gilbreath's conjecture survives Cramér random primes","Prime-difference conjecture holds in probabilistic model","Gilbreath's conjecture fails only at structured zero blocks","Random prime gaps satisfy Gilbreath's difference conjecture","Tower argument shows Gilbreath's conjecture holds probabilistically"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The deterministic inverse theorem assumes a Cramér-type upper bound on prime gaps (itself unproved) and additionally requires ruling out long zero-blocks and long shallow two-valued blocks in the array — conditions the authors note look difficult to establish even assuming the Hardy-Littlewood prime tuples conjecture.","fun_headline_variants_meta":{"raw":{"variants":["Gilbreath's conjecture survives Cramér random primes","Prime-difference conjecture holds in probabilistic model","Gilbreath's conjecture fails only at structured zero blocks","Random prime gaps satisfy Gilbreath's difference conjecture","Tower argument shows Gilbreath's conjecture holds probabilistically"]},"model":"glm-5.2","effort":"low","cost_usd":0.0,"raw_usage":{"total_tokens":589,"prompt_tokens":523,"completion_tokens":66,"prompt_tokens_details":null},"tokens_in":523,"tokens_out":66,"duration_ms":27984,"temperature":1.0,"reasoning_tokens":null,"cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-10T02:34:28.624429+00:00","model_set":{"reader":"glm-5.2"},"falsifier":"If normalized prime gaps were shown to concentrate in some 2-separated set (violating axiom (ii) of Theorem 1.3), or if long blocks of zeroes or long shallow {0,d}-valued blocks were shown to occur in the prime gap Gilbreath array, the path to proving the original conjecture via this paper's framework would be blocked.","supporting_citations":[],"review_version":2}