\section{Introduction} How many positive integers \(n\le X\) have the property that both \(n\) and \(n+1\) are sums of two squares? Write \(b_d(n)\) for the indicator of the positive values of \(x^2+dy^2\), with \(x,y\in\Z\), and put \(b_d(n)=0\) for \(n\le0\). Coordinates are unrestricted and may vanish; each numerical value is counted once. Under the analytic inputs stated below, we obtain \begin{equation}\label{eq:ck-headline} \#\{1\le n\le X:b_1(n)=b_1(n+1)=1\} \sim\frac{X}{2\log X}. \end{equation} This is the shift-one Connors--Keating prediction. It belongs to Smilansky's five-form family of pair-correlation predictions. \begin{corollary}\label{cor:smilansky-five-pairs} Under the inputs of Section~\ref{sec:inputs}, for \(d\in\{1,2,3,4,7\}\) and fixed \(h\ne0\), \[ \sum_{1\le n\le X}b_d(n)b_d(n+h) =\mathscr T_{d,h}\frac X{\log X}+o_{d,h}(X/\log X), \] with precisely the coefficients predicted in \cite[Section 6.3, equations (6.2)--(6.4)]{smilansky}. In particular, \begin{equation}\label{eq:minus-eight-shift-one} \sum_{n\le X}b_2(n)b_2(n+1) \sim\frac1{\sqrt2}\frac X{\log X}. \end{equation} \end{corollary} Section~\ref{sec:quadratic-pairs} evaluates the coefficients and exceptional-prime conditions. \subsection{A second headline: genus governs primitive value counts} Fix one negative fundamental discriminant \(D\) and two fixed discriminants \(\Delta_i=f_i^2D\). Let \(Q_i,Q'_i\) be primitive positive definite integral binary forms of discriminant \(\Delta_i\), in the same genus for each \(i=1,2\), in the one field \(\Q(\sqrt D)\). If \(P_Q\) is the set of positive values of \(Q\) at coprime coordinates and \(R_{Q_1,Q_2}(N)=\#\{(a,b)\in P_{Q_1}\times P_{Q_2}:a+b=N\}\), then the classical route gives \begin{equation}\label{eq:genus-headline} R_{Q_1,Q_2}(N)=R_{Q'_1,Q'_2}(N) \left(1+O((\log N)^{-1/5})\right) \end{equation} on sufficiently large common locally admissible targets. The slots may have different genera and conductors. Under the recent analytic inputs the same count has an explicit leading term of order \(NJ_S(N)/\log N\), with relative error \(o(1)\). No class-number divisor occurs: a value can belong to every class in its compatible genus. The common-field theorem and its principal specialization are in Section~\ref{sec:forms}. The counting conventions and strengths are different: \begin{center}\small \begin{tabular}{@{}>{\raggedright\arraybackslash}p{.19\linewidth}>{\raggedright\arraybackslash}p{.34\linewidth}>{\raggedright\arraybackslash}p{.41\linewidth}@{}} \hline Result family&What is counted&Conclusion\\\hline CK/Smilansky&Ordinary values; unrestricted coordinates&Fixed-shift asymptotic with the stated little-\(o\) error\\[1mm] Subgroup/coset sums&Integers with prescribed prime support&Target-uniform asymptotic with a logarithmic saving\\[1mm] Primitive-form sums&Values admitting coprime coordinates in two fixed orders of one field&Explicit leading term with relative \(o(1)\)\\[1mm] Genus replacement&Comparison of two primitive-form value counts&Relative logarithmic saving through the classical route\\\hline \end{tabular} \end{center} The first three asymptotic families use the imports in Section~\ref{sec:inputs}; genus replacement does not. A \emph{primitive form} \([a,b,c]\) has \(\gcd(a,b,c)=1\). A \emph{primitive representation} \(Q(x,y)\) has \(\gcd(x,y)=1\). \emph{Coprime summands} in \(a+b=N\) means \(\gcd(a,b)=1\), a separate condition imposed only when stated. For example, \(4=2^2+0^2\) is an ordinary sum of two squares, but any representation of four has both coordinates even and is not primitive. Consequently the primitive-form theorem does not extend the ordinary CK statement to arbitrary forms. Its class-error saving also does not supply a logarithmic remainder for the full conductor-prefix leading term. \subsection{The common mechanism, in the CK example} The central result is the indicator-to-model comparison of Theorem~\ref{thm:model-comparison}. Multiplication by suitable primes preserves an indicator and its arithmetic model. One-variable means center their difference, and a positive sparse sieve controls exceptional prime-factor records. A shared graph reduction then leaves two frequency estimates: one for fixed shifts and one for reflected targets. Their assembly replaces the indicators in the binary count by the models. The model correlation can then be evaluated from local arithmetic. Here is the model for \(b_1\). At outer scale \(X\), put \(L=\log X\), \(y=\exp(L^{2/5})\), \(\mathcal P_y=\{p:p\le y\}\), and \(V_y=\prod_{p\le y}(1-1/p)\). Write every \(n\) uniquely as \(n=rv\), where \(r=n_{\mathcal P_y}\) contains its full small-prime powers and \(v\) has no prime factor at most \(y\). The rough ordinary-value series is \[ F_y(s)=\prod_{\substack{p>y\\p\equiv1\ (4)}}(1-p^{-s})^{-1} \prod_{\substack{p>y\\p\equiv3\ (4)}}(1-p^{-2s})^{-1} =(s-1)^{-1/2}H_y(s). \] For a sufficiently deep fixed \(K\), its smooth density expansion is \[ M(v)=\sum_{j=0}^K\frac{H_y^{(j)}(1)}{j!} \frac{(\log v)^{-1/2-j}}{\Gamma(1/2-j)},\qquad \mathcal D(rv)=b_1(r)\bigl(1+\chi_{-4}(v)\bigr)\frac{M(v)}{V_y}. \] These formulas apply on the polynomial rough scales used in the proof; the smooth bounded extension is specified in Section~\ref{sec:model-preparation} and the quadratic application. The leading term of \(M\) is \(H_y(1)/\sqrt{\pi\log v}\). Dividing by \(V_y\), the density of rough integers, gives the conditional density on the rough variable. Every allowed rough value has \(\chi_{-4}(v)=1\). The correction \(1+\chi_{-4}(v)\) retains this bias; average density alone would miss a leading contribution. For a split prime \(p\le y\), multiplication by \(p\) changes only \(r\), preserves \(b_1(r)\), and leaves both \(v\) and \(\mathcal D\) unchanged. Replacing \(b_1(n)b_1(n+h)\) by \(\mathcal D(n)\mathcal D(n+h)\) with an error \(o(X/\log X)\) is the analytic task. The remaining model product gives the coupled local factors and the coefficient \(1/2\) in \eqref{eq:ck-headline}. \subsection{Semigroups and the reflected-target obstacle} The same mechanism reaches a different geometry. Let \(C_{12}=\{n\ge1:p\mid n\Rightarrow p\equiv1\pmod{12}\}\), including one. Every sufficiently large \(N\equiv2\pmod{12}\) is a sum of two elements of \(C_{12}\), and the ordered value-pair count \(R_{12}(N)=\#\{(a,b)\in C_{12}^2:a+b=N\}\) satisfies \begin{equation}\label{eq:opening} R_{12}(N)=A_{12}J_{12}(N)\frac{N}{(\log N)^{3/2}} \left(1+O((\log N)^{-\delta})\right), \qquad\delta>0. \end{equation} The positive factors are explicit in Theorem~\ref{thm:c12}. For \(C^-_4=\{n\ge1:p\mid n\Rightarrow p\equiv3\pmod4\}\), every sufficiently large even target likewise has an explicit asymptotic of order \(N/\log N\), with a positive logarithmic saving. The subgroup and coset theorems allow fixed arithmetic data and retain all surviving characters, including the complex branches forced by the prime class \(3\pmod{10}\). Numerical saving exponents are collected in Section~\ref{sec:assembly}. One-variable density does not control the particular line \(a+b=N\), and averages can conceal exceptional targets. The nonconstant linear parts of \(n\) and \(N-n\) are dependent, so the independent-linear-form correlation theorems of Loughran--Matthiesen do not directly apply \cite[Theorems 1.9 and 3.10]{loughranmatthiesen}. The reflected frequency argument must retain cancellation and normalization when the target enters the rational denominators. Iwaniec--Pomyka\l a's quartic-norm work is a direct antecedent \cite{iwaniecpomykala}. For \(\Q(\zeta_{12})\), its splitting conditions select the primes \(1\pmod{12}\). Their Theorem~2 gives conditional pointwise order bounds for coprime sums; hypothesis (P4) is uniform in growing shifts. Their Theorem~1 treats fixed differences unconditionally. They count numerical values once. Our unrestricted leading term gives a coprime asymptotic by M\"obius inversion, with the extra target factor in Corollary~\ref{cor:coprime-c12}. \subsection{Attribution and dependencies} Walsh's multiplicative-invariance framework treats positive integer divisor orders and fixed nonzero shifts \cite{walsh}. His Section~1.2.4, pp.~5--6, proposes sums-of-two-squares indicator correlations. Here we supply their complete character models, arithmetic hypothesis checks and local factors, and the uniform reflected-target argument. The recent inputs are the weighted-dilation root-moment theorem \cite[Section 3]{weightedgraphs} and the all-conductor zero-free half-plane \(\Re s>11/12\) \cite[Part I, Theorem 3.1]{zerofree}; Part II is not used. Walsh notes a possible classical recovery of his fixed-shift theorem \cite[p.~2]{walsh}, but his printed Lemma~5.1 and Section~6.4 retain that zero-free input. Our polynomial-logarithmic fixed-shift moduli still occur at Mellin heights up to \(x(\log x)^C\). Section~\ref{sec:fixed-classical-boundary} proves the classical intermediate-height estimate and states the remaining high-frequency obligation. Both recent inputs remain in our fixed-shift result; such a replacement would also leave the reflected growing-conductor problem. Appendix~\ref{sec:older-route} proves classical quadratic order bounds by Kloosterman completion and positive restoration. With the class-error sieve these give genus comparison independently of the recent inputs. All data are fixed; thresholds are not numerical. The companion paper \cite{exactcompanion} gives exact classifications and finite certificates for four primitive quadratic norm sumsets. The deposited window paper concerns the placement of summands. \subsection{Notation} \begin{center}\small \begin{tabular}{@{}>{\raggedright\arraybackslash}p{.28\linewidth}>{\raggedright\arraybackslash}p{.66\linewidth}@{}} \hline Symbol&Role\\\hline \(X\), \(L=\log X\), \(h\)&Outer scale, its logarithm, and the fixed shift\\ \(\mathcal P_y\), \(n_{\mathcal P_y}\), \(V_y\)&Small primes, complete small-prime part, and rough-integer density\\ \(f_i,\mathcal D_i,\xi_i\)&Indicators, arithmetic models, and centered errors\\ \(D,D_i\), \(U_i\)&Auxiliary prime products and their dyadic lengths\\ \(H_i\), \(\mathscr A\), \(\mathcal E\)&Residue subgroups, a prime coset, and forbidden prime sets\\ \(B_0\), \(S\)&Finite exceptional-prime sets in the ordinary-indicator and primitive-form applications\\ \(\Delta_i=f_i^2D\)&Order discriminants sharing one fixed fundamental discriminant \(D\); here \(f_i\) denotes a conductor\\ \(P_Q,\Lambda_Q\), \(G\)&Primitive numerical values, their local shadow, and the applicable class group\\\hline \end{tabular} \end{center} The modulus and the arithmetic data stay fixed. Definitions in each application specify its groups and exceptional primes; \(h\) is reserved for the additive shift throughout.