Route107 triage: repair the object and analytic mechanism before fitting residues Recommendation: continue only with a bounded analytic reformulation, not with the proposed ordinary Euler-product/pole-at1 argument. The corrected H log^2 H law and its constants remain unproved here. The following obstructions concern the stated formulation and method, not the broad route. 1. What the actual submitted script measures ------------------------------------------- Let A=2C2 and Q(H)=sum_{0<|h|=1 the baseline weights are exactly W_all = sum_{0<|h|=5,p|h} (p-2)/(p-4) prod_{p>=5,p|h^2-4} (p-3)/(p-4), C=prod_{p>=5} p(p-4)/(p-2)^2. The normalized prime2 factor is2 on even h and0 on odd h. The normalized prime3 factor is3 on multiples of3 and0 elsewhere. For p>=5 the generic factor is p(p-4)/(p-2)^2; coincidences h=0 modp multiply it by(p-2)/(p-4), and h=+/-2 modp multiply it by(p-3)/(p-4). These cases are disjoint. Even after removing the6-support and common constant, the coefficients are not multiplicative. Set F(n)=a(6n)/(6C). Then, exactly, F(1)=1, F(2)=8/3, F(3)=2, F(6)=224/195. Since gcd(2,3)=1, F(6) != F(2)F(3). Hence the ordinary Dirichlet series does not acquire an ordinary Euler product just because each a(h) has a product over primes. Dependence on h-2 and h+2 is essential. A Dirichlet series and Perron formula may still be used, but their continuation and pole structure need a new argument for shifted correlations. 3. Elementary leading mean, not the desired secondary terms ---------------------------------------------------------- The local identity (1/p) sum_{h modp} a_p(h)=1 follows either from the displayed factors or from autocorrelating the indicator of the twin-allowed residues modulo p. By CRT it holds for every fixed finite prime product. Exchanging finite-sieve and H limits without uniformity is not justified by that identity alone. There is also an elementary positive divisor expansion giving the leading mean. Expand the factors as1+2/(p-4) for p|h and1+1/(p-4) for p|h-2 or p|h+2. Write alpha(d)=prod_{p|d}2/(p-4) and beta(d)=prod_{p|d}1/(p-4), supported on squarefree d coprime to6. The resulting triples d0,d-,d+ are pairwise coprime; incompatible triples contribute zero. Together with6|h they specify one residue class modulo6D, D=d0*d-*d+, by CRT. For any residue class modulo L, sum_{1<=h=1}(a(h)-1)h^(-s), initially in its half-plane of absolute convergence. The triangular Perron kernel for the all-shift centered sum, divided by A^2, is 2/(2*pi*i) integral B(s) H^(s+1)/[s(s+1)] ds. Thus a cubic pole of B at s=1 would produce H^2 times a quadratic polynomial in log H, not H log^2 H. The desired secondary term would come from a cubic pole of the INTEGRAND at s=0; because the kernel already has1/s, a double pole of B at0 is the corresponding candidate. This is pole bookkeeping, not proof that such continuation/poles exist. 5. Primary-source hypothesis map ------------------------------- Montgomery--Soundararajan, Primes in short intervals, arXiv math/0409258, https://arxiv.org/html/math/0409258: Theorem2 and equations(8),(17) sum over all independently varying distinct coordinates in a box. Equation(17) counts ordered tuples, with every k-element set appearing k! times; the search synthesis incorrectly described it as unordered and mislabeled the theorem. Section2 Lemma4, equations(47)--(49), treats the ordinary two-offset series S({0,m}), not two translated fixed twin pairs. I read those primary statements, replacing the prior abstract/secondary-only lookup. Kuperberg, Sums of singular series along arithmetic progressions and with smooth weights, arXiv2301.06095, published IJNT2025, DOI10.1142/S1793042125500046: https://arxiv.org/html/2301.06095 Definition(4) and Theorem1.2 fix congruence classes modulo a fixed r but still sum independently varying coordinates. Definition(10) and Theorems1.3/1.5 use products of fixed one-variable smooth weights with compact support and Fourier decay. They do not impose the exact equalities d2=d1+2 and d4=d3+2. Taking r or delta-like weights to depend on H would require uniform estimates not furnished by those statements. Also their centered S0 is an inclusion-exclusion over all subsets, not simply S4-A^2. These are useful mechanisms, not an off-the-shelf proof of the fixed-pair secondary asymptotic. No exact matching theorem was located in this bounded update. That is not proof of novelty. I did not read the entire thesis or assert that no other chapter or paper can contain a relevant result. 6. Bounded next experiment ------------------------- First standardize D(H), including the treatment of h=0,+/-2. Then derive the above shifted-divisor/CRT expansion with an explicit cutoff and an error budget. Separate the density-product tail from the weighted CRT boundary contribution. The gate is an analytic continuation/error statement capable of resolving terms of size H, not another numerical fit or an ordinary multiplicative Euler product. A bounded first pass should either identify such a cancellation lemma and map a primary theorem to it, or isolate the exact weighted three-divisor boundary sum whose available bound is too large. The full a,b,c and the proposed o(H) remain open. The variance interpretation still needs a Hardy--Littlewood conjecture uniform in the shifts and an error summable against the triangular weights; pointwise k-tuple asymptotics alone do not provide that. The accompanying small rational checker exercises local factors, primewise means, parity/centering identities, the uniform progression control and the exact nonmultiplicativity witness. It does not rerun the published million-scale table or infer its limiting constants.