{"id":"96b72182-82b3-4cce-ae14-9e4537d9f30b","arxiv_id":"2506.03005","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Prime patterns in blocks of 8, 64, and 512 integers are encoded as bytes, and the paper gives empirical histograms and binomial-style estimates for their counts.","lead":"This paper encodes blocks of eight integers into byte-sized codes, then stacks the codes into a multi-scale tree that shows where primes occur within stretches of 8, 64, and 512 numbers. It reports that, for large ranges, the histograms of these codes settle into a stable shape, and it gives approximate formulas with correction factors fitted to the data.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Observation 4's estimator depends only on Hamming weight j, so it cannot estimate individual histogram bins; the claimed invariant shape is left without support.","rationale":"The reader identified the independence assumption as the weakest point; the present concern is adjacent but sharper. Even if the Bernoulli independence assumptions were granted, the formula in Observation 4(2) predicts identical counts for all k-patterns with the same number of 1-bits, because the correction factor C_j is indexed only by j. Table 3 contradicts this prediction at level 3 with highly significant within-class variation. Therefore the claimed estimation for each value in the histogram is not actually per-value, and the invariant-shape conjecture cannot be supported by the stated model. The paper remains valuable as an exploratory encoding and visualization tool: Observation 1 is an elementary but genuine fact, the multi-scale construction is clearly defined, and the histograms are reproducible from public prime tables. However, the central quantitative claim needs either a per-pattern correction model or a clear restriction to Hamming-weight aggregates, plus code-backed validation. The reader's CONDITIONAL verdict remains appropriate, hence UNCHANGED.","tokens_in":12218,"tokens_out":7734,"duration_ms":94844,"concrete_test":"Using the public first-fifty-million-primes files, recompute Table 3 at m = Nmax and at m = Nmax/2. For each Hamming-weight class j, fit the single factor C_j(m) by maximum likelihood from the aggregate count of all patterns with np_k(n) = j, then compute Pearson residuals r_n = (π_m(n) − E_n)/sqrt(E_n) for each individual pattern n, with E_n = C_j(m)(m/8^k) q_{k-1}(m)^j (1−q_{k-1}(m))^{8−j}. If a chi-square test over patterns within a fixed j is wildly inconsistent with sampling noise, the estimator is not bin-level and cannot support the invariant-shape claim. As a second check, compare the ratio π_{Nmax}(n)/π_{Nmax}(n′) for two patterns with the same Hamming weight against the same ratio at m = Nmax/2; instability of these within-class ratios between cutoffs would directly undermine invariance.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing step is the move from Observation 4(2), which gives π_m^(k)(n) ≈ C_j^(k)(m) (m/8^k) q_{k-1}(m)^j (1−q_{k-1}(m))^{8−j} with j = np_k(n), to the claim that one has an estimation for each value in the histogram. The right-hand side depends on n only through np_k(n), so it assigns the same expected count to every k-pattern with the same number of 1-bits; the correction factor C_j^(k)(m) is indexed by j only and cannot remove this degeneracy. Table 3 shows the degeneracy is empirically false at level 3: among j=7 patterns, counts range from 35,878 (pattern 127) to 36,182 (pattern 253); among j=6 patterns, counts range from 649 (pattern 63) to 868 (pattern 126). These differences are far larger than Poisson counting error at about 1.9 million patterns, where σ ≈ 190 for the j=7 class and σ ≈ 28 for the j=6 class. Thus the formula describes, at best, a 9-bin Hamming-weight aggregate, not the 256-bin histogram whose irregular but repeating shape is the paper's central observation. Consequently, the invariant-shape conjecture in Section 2.2 is not derivable from, and is not even compatible with, the stated estimator; the irregular within-weight structure visible in Figure 8a and Table 3 is unmodeled. The central claim therefore rests on an unexplained empirical regularity, not on Observation 4 as written.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a multi-scale binary encoding of integer sequences, in which blocks of 8 integers are represented by 8-bit patterns, and the construction is iterated to higher scales. The method is applied to prime numbers up to 982,451,200, producing level-1 patterns restricted to 14 values, level-2 and level-3 histograms, and a conjecture that these histograms share an invariant shape for sufficiently large intervals. The paper also gives approximate formulas for histogram bin counts using binomial probabilities multiplied by correction coefficients C_j^{(k)}(m), and applies the same encoding to Mersenne primes. The main claimed contributions are the multi-scale representation itself, the observation of an invariant histogram shape, and 'an estimation for each value in the histogram and at all scales.'","tokens_in":12495,"tokens_out":9242,"duration_ms":104743,"significance":"The descriptive parts of the paper are clear and easy to verify from public prime tables: Observation 1 correctly identifies the 14 admissible level-1 patterns, and Tables 1-3 report reproducible histogram data. The multi-scale encoding is a reasonable exploratory tool for visualizing prime distributions at different granularities, and the observed repetition of histogram shapes across scales is a potentially interesting phenomenological fact. However, the paper's analytic claims are not established: the correction coefficients are fitted to the same data they are supposed to predict, and the level-2 and level-3 formulas depend only on the Hamming weight of a pattern, so they cannot estimate individual histogram bins. The invariant-shape conjecture is presented as an empirical observation but is not supported by the proposed model, and no quantitative or out-of-sample evidence is supplied. The paper is honest in stating that it does not claim new results on primes, but the central 'estimation' claim is currently overstated.","major_comments":[{"comment":"The estimator in Observation 4(2), namely pi_m^(k)(n) approximately equal to C_j^(k)(m) (m/8^k) q_{k-1}(m)^j (1 - q_{k-1}(m))^{8-j} with j = np_k(n), depends on n only through the Hamming weight j. It therefore assigns the same expected count to every k-pattern with the same number of ones, so it cannot provide 'an estimation for each value in the histogram' as claimed in the abstract. Table 3 shows the degeneracy is empirically false at level 3: among j=6 patterns, counts range from 649 (pattern 63) to 868 (pattern 126), a spread of 219 against a Poisson standard deviation of about 28 for counts near 800; among j=7 patterns, counts range from 35,878 (pattern 127) to 36,182 (pattern 253). The formula at best describes a 9-bin Hamming-weight aggregate, not the 256-bin histogram whose irregular shape is the paper's central observation. The invariant-shape conjecture in Section 2.2 is therefore not derivable from, and is not even compatible with, the stated estimator.","section":"Section 2.3, Observation 4(2) and Table 3"},{"comment":"Equations (8) and Observation 3(2) are not independent predictions. The correction coefficients C_j^(1)(m) and C_j^(2)(m) are introduced after the empirical histograms are displayed, and the text states that they were introduced 'to correct the approximations and the independence assumption'; no closed-form expression, out-of-sample validation, or error analysis is provided. Once the fitted C_j are inserted, the level-1 formula for a specific pattern is essentially a restatement of the measured count, and the level-2 formula fits only Hamming-weight totals. The binomial factors also rest on the independence of the four odd numbers in a block and of adjacent blocks, which is known to fail for primes; twin and k-tuple correlations are absorbed into the fitted C_j without a model. Consequently, the claim that the paper proposes a mathematical estimation of each histogram bin is not supported.","section":"Section 2.1, Equation (8), and Section 2.2, Observation 3"},{"comment":"The conjecture that all histograms of f^(k)(n) for k >= 2 share the same invariant shape is supported only by visual inspection of histograms computed on the single interval [1, Nmax] and by Observation 4, which, as noted above, depends only on Hamming weight. No evidence is given that the shape is independent of the upper limit m beyond Nmax, no quantitative measure of shape invariance (for example, a normalized distance between histograms at different m) is provided, and the irregular within-weight structure visible in Figure 8a and Table 3 is not explained by the proposed formula. The invariant-shape claim is therefore an unexplained empirical regularity rather than a consequence of the model.","section":"Section 2.2, The Invariance Shape of the k-Patterns Histogram"}],"minor_comments":[{"comment":"In the paragraph preceding Algorithm 1, the text says 'it determines L1 = D2B(f^(2)(8(8n + j) + i))', but this should refer to f^(1), since L1 is a level-1 pattern; the algorithm itself correctly uses pattern1[k].","section":"Section 3, Reconstruction Algorithm"},{"comment":"Observation 1 states that f^(1)(n) takes only the 14 values in P1, but the value 106 occurs only for n = 0; for n >= 1 it is impossible. The statement should specify the domain n = 0 separately to avoid ambiguity.","section":"Section 2.1, Equation (5)"},{"comment":"Definition 2 says the primes are dense 'if for all n' the intersection is non-empty, but for n such that P^(k)(n) lies entirely beyond m the intersection is empty by construction; the definition should restrict n to blocks intersecting [1, m].","section":"Section 2.2, Observation 5 and Definition 2"},{"comment":"Observations 6 and 7 are based on finite computations (1 < m <= M with M = 1000, and 0 <= m <= 500) but are phrased as general statements; the finite scope should be stated explicitly in each observation.","section":"Section 4, Observations 6 and 7"},{"comment":"The column header 'n' in the tables of Figure 7 denotes the pattern value, not the index n used elsewhere in the paper; renaming this column would prevent confusion.","section":"Figure 7"},{"comment":"Reference [2] contains a typo ('Elsvier' for 'Elsevier'), and the data source in reference [6] cites a Wikipedia page; a direct citation to the PrimePages data files would be more appropriate.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is better characterized as an exploratory, empirically driven study than as a mathematical derivation. The level-1 analysis (Observation 1 and Table 1) is solid, and the multi-scale encoding is a nice descriptive tool. However, the central estimation formulas for levels 2 and 3 are not valid as statements about individual histogram bins, and the correction coefficients are fitted to the data rather than derived. A revision that explicitly reframes the contribution as a descriptive observation, adds out-of-sample tests, and either removes or substantially weakens the 'estimation for each value' claim could make the paper acceptable for an applied or experimental venue. In its current form, the abstract and Observations 3-4 overstate what is established."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague, the genuinely new piece here is the nested 8-bit encoding f^(k) and the observation that level-1 patterns for primes take only 14 values. That part is easy to verify and correct. The paper is also honest in its abstract: it claims no breakthroughs. But the statistical core, Observations 3 and 4, has a problem the authors don't flag: the estimator depends on n only through the Hamming weight j = np_k(n), so it assigns the same expected count to every k-pattern with the same number of 1-bits. At level 1 this is harmless because the 14 allowed patterns have distinct weights. At level 2 and above it is not. Table 3 shows the degeneracy is empirically false: among the j=7 patterns at level 3, counts range from 35,878 to 36,182, and among j=6 patterns from 649 to 868, both far outside Poisson noise. The formula therefore describes a 9-bin Hamming-weight aggregate, not the 256-bin histogram whose irregular shape is the paper's central observation. The invariant-shape conjecture is not supported by Observation 4 as written; it rests on an unmodeled empirical regularity.\n\nThe correction coefficients C_j(m) are fitted to the same histograms they estimate, so Equation (8) is calibration, not derivation. That is a soft spot, but the paper does call them corrections and says formal determination is future work, so it is a minor issue compared with the Hamming-weight degeneracy. There is no code or data shipped, though the method is reproducible from the public prime list.\n\nWhat is genuinely good: the encoding scheme is clean, the 14-pattern observation is a real arithmetic constraint, and the reconstruction algorithm works. The Mersenne section is thin—five simple observations—but it doesn't hurt the main narrative.\n\nWho is this for? Anyone building visual or query tools for prime distributions might get use from the representation. A number theorist will not learn new mathematics. I would send it to peer review, because the representation is new and verifiable, but the referee should insist the estimation/invariance claims be weakened to Hamming-weight aggregates or properly supported. The descriptive part deserves to be published; the statistical part, as it stands, does not.","headline":"A clean new visual encoding for primes, with a histogram estimator that only fits Hamming-weight aggregates, not the 256 bins it claims.","tokens_in":13092,"tokens_out":2434,"would_cite":false,"duration_ms":25380,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11A41","11N05","11Y16"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proposes a multi-scale 8-bit encoding of primality and claims that, for large intervals, all pattern histograms converge to one invariant shape, with an explicit binomial estimate for every bin.","keywords":["multi-scale analysis","prime numbers","8-bit patterns","invariant histogram shape","binomial model","Mersenne primes","hierarchical encoding","density of primes"],"falsifier":"Compute the histograms of $f^{(2)}(n)$ and $f^{(3)}(n)$ over much larger intervals, for example up to $m=10^{11}$, and check whether the empirical ratios $\\pi_m^{(k)}(j)/\\big((m/8^k)\\,q_{k-1}(m)^j\\,(1-q_{k-1}(m))^{8-j}\\big)$ stabilize inside a fixed bounded band as $m$ grows, and whether the normalized histograms keep the same outline as the one reported near $m=10^9$. If the ratios drift without bound or the outline changes, the invariant-shape conjecture and the binomial estimate are refuted.","tokens_in":11929,"feed_emoji":"🔢","tokens_out":15992,"duration_ms":139981,"temperature":0.7,"pith_summary":"The paper introduces a hierarchical way to encode an integer set: at scale $k$, blocks of $8^k$ consecutive integers are compressed into an 8-bit pattern, written as a decimal between 0 and 255, that records where the property holds. Applied to primality, only 14 of the 256 possible finest patterns ever occur, and the paper proposes that for large enough intervals the histograms of these patterns at every coarser level settle into one invariant shape, characteristic of prime numbers. The paper also derives an explicit estimate for each histogram bin, a binomial-type formula with empirical correction factors built on the probability $p(m)=2\\,\\mathrm{li}(m)/m$ that an odd number below $m$ is prime. If the claim holds, the method gives a compact, scale-independent statistical signature of primality, and the same machinery is shown to sharply restrict the patterns of Mersenne primes.","feed_headline":"Prime histograms settle into one shape at every scale","feed_subtitle":"A binomial-style formula predicts how often each 8-bit prime pattern occurs, once the interval is large enough.","key_machinery":"The central object is the nested sequence of 8-bit patterns $f^{(k)}(n)$, defined recursively by $f^{(1)}(n)=\\sum_{j=0}^{7}\\phi_0(8n+1+j)\\,2^{7-j}$ and $f^{(k+1)}(n)=\\sum_{j=0}^{7}\\phi_k(f^{(k)}(8n+j))\\,2^{7-j}$, where each $\\phi_k$ flags whether a sub-block contains at least one prime. The argument rides on treating the number of ones in a $k$-pattern as a binomial count with per-trial probability $q_{k-1}(m)$ (the chance that a $(k-1)$-pattern contains a prime), together with empirical correction factors $C_j^{(k)}(m)$ that absorb the failure of independence and the approximation in $p(m)=2\\,\\mathrm{li}(m)/m$.","core_discovery":"Stated on the paper's own terms: for primes, the multi-scale sequences $f^{(k)}(n)$ take values in $\\{0,\\dots,255\\}$, and for a sufficiently large interval $[1,m]$ the histograms of the $k$-patterns, for every $k\\ge 2$, share the same outline, a shape that remains invariant as $m$ grows and that the paper treats as a characteristic of prime numbers. The count of each histogram bin is approximated by $C_j^{(k)}(m)\\,\\frac{m}{8^k}\\,q_{k-1}(m)^{j}\\,(1-q_{k-1}(m))^{8-j}$, where $j$ is the number of ones in the pattern, $q_{k-1}(m)=1-(1-p(m))^{4\\cdot 8^{k-2}}$, and $p(m)=2\\,\\mathrm{li}(m)/m$; the empirical factors $C_j^{(k)}(m)$ are bounded (below 2 for the tested levels, and inside $[1/2,3/2]$ for the first-level factors when $m>16{,}777{,}216$) and absorb the correlations that a strict independence model misses. At the finest scale the admissible patterns are exactly the 14 masks compatible with divisibility by 3, and for Mersenne primes the first-level patterns reduce to $\\{0,2,8,10,42,128\\}$, with no observed block containing more than one Mersenne prime.","pith_inferences":["If the invariant shape survives tests at much larger $m$, the same hierarchical encoding could be applied to other arithmetic properties by replacing $\\phi_0$, with the binomial model adjusted by an appropriate per-block probability.","The approximate symmetries among correction factors (for instance $C_2\\approx C_{128}$) suggest a reflection structure in how residues modulo 8 correlate with primality; checking whether these symmetries persist to larger scales would probe those correlations directly.","The method could serve as a statistical test of pseudorandomness: a set whose multi-scale histograms deviate from the prime signature would be detectably non-random in a way that ordinary density estimates would miss.","One testable extension is to compute the ratios $\\pi_m^{(k)}(j)/\\big((m/8^k)\\,q_{k-1}(m)^j\\,(1-q_{k-1}(m))^{8-j}\\big)$ for several $k$ over disjoint million-scale windows; stability of these ratios would confirm the conjecture and produce a practical calibration of the correction factors."],"forward_implications":["The invariant histogram shape provides a scale-independent statistical signature of primality: the same outline appears at level 2, level 3, and beyond once $m$ is large enough.","The binomial estimate gives a way to predict how often each local prime pattern occurs without sieving the whole interval, up to the bounded correction factors.","The 14-pattern restriction at level 1, together with the reconstruction algorithm, allows a large interval of primes to be stored and queried through a short sequence of small integers.","For Mersenne primes the pattern set shrinks to $\\{0,2,8,10,42,128\\}$, and the observation that no block contains more than one Mersenne prime quantifies their extreme rarity over the tested range.","The density argument shows there is a finite scale at which every block of size $8^k$ can no longer contain a prime, with a practical bound around $m=e^{27}$ for level 3."],"supporting_citations":[{"why":"The archived table of the first fifty million primes; every empirical histogram, correction factor, and symmetry observation in the paper is computed against this dataset.","marker":"[6]"},{"why":"Companion documentation of that prime dataset, cited together with [6] as the source of primes below 982,451,653.","marker":"[7]"},{"why":"Supplies the density fact behind Observation 5: no regular partition of the integers can guarantee a prime in every part, motivating the finite-scale density bound.","marker":"[8]"}],"fun_headline_variants":["Prime patterns show same histogram shape at every scale","One histogram shape for primes at all zoom levels","Prime histograms converge to a universal shape","All scales of prime patterns share one histogram outline"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole histogram estimate assumes that every odd number in $[1,m]$ has the same probability $p(m)=2\\,\\mathrm{li}(m)/m$ of being prime and that the four odd candidates in each 8-block, and the blocks across the interval, behave like independent coin flips; real primes are correlated, so the correction factors $C_j^{(k)}(m)$ carry the entire burden of fixing that assumption numerically.","fun_headline_variants_meta":{"raw":{"variants":["Prime patterns show same histogram shape at every scale","One histogram shape for primes at all zoom levels","Prime histograms converge to a universal shape","All scales of prime patterns share one histogram outline"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000595,"raw_usage":{"total_tokens":2853,"prompt_tokens":1079,"completion_tokens":1774,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":695,"completion_tokens_details":{"reasoning_tokens":1716}},"tokens_in":695,"tokens_out":1774,"duration_ms":12025,"temperature":1.0,"reasoning_tokens":1716,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T11:11:56.212131+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the histograms of $f^{(2)}(n)$ and $f^{(3)}(n)$ over much larger intervals, for example up to $m=10^{11}$, and check whether the empirical ratios $\\pi_m^{(k)}(j)/\\big((m/8^k)\\,q_{k-1}(m)^j\\,(1-q_{k-1}(m))^{8-j}\\big)$ stabilize inside a fixed bounded band as $m$ grows, and whether the normalized histograms keep the same outline as the one reported near $m=10^9$. If the ratios drift without bound or the outline changes, the invariant-shape conjecture and the binomial estimate are refuted.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The archived table of the first fifty million primes; every empirical histogram, correction factor, and symmetry observation in the paper is computed against this dataset."},{"cited_title":"Caldwell and Jr., G","cited_arxiv_id":null,"evidence_quote":"Companion documentation of that prime dataset, cited together with [6] as the source of primes below 982,451,653."},{"cited_title":"Tenenbaum, Introduction to Analytic and Probabilistic Number Theory, Cambridge Studies in Advanced Mathematics, 1995","cited_arxiv_id":null,"evidence_quote":"Supplies the density fact behind Observation 5: no regular partition of the integers can guarantee a prime in every part, motivating the finite-scale density bound."}],"review_version":1}