\section{Primitive forms and invariance within a genus}\label{sec:forms} Fix a negative fundamental discriminant \(D\), put \(K=\Q(\sqrt D)\), and let \[ \mathcal O_i=\Z+f_i\mathcal O_K,\qquad \Delta_i=f_i^2D,\qquad G_i=\operatorname{Pic}(\mathcal O_i) \quad(i=1,2). \] Let \(Q_i\) be a primitive positive definite integral binary form of discriminant \(\Delta_i\). For any such form \(Q\), define \[ \begin{aligned} P_Q&=\{Q(x,y):x,y\in\Z,\ \gcd(x,y)=1\},\\ \mathscr V_{p,Q}&=Q(\Z_p^2\setminus p\Z_p^2),\qquad \Lambda_Q=\{n\ge1:n\in\mathscr V_{p,Q}\text{ for every }p\}. \end{aligned} \] Only positive values are included. Coordinates may vanish, and each numerical value is counted once. Write \[ R_{T_1,T_2}(N)=\#\{(a,b)\in T_1\times T_2:a+b=N\},\qquad R_{Q_1,Q_2}=R_{P_{Q_1},P_{Q_2}}. \] Two forms are in the same genus when they are equivalent over every \(\Z_p\) and over \(\mathbb R\). These equivalences preserve primitive vectors and therefore the primitive local value sets. Put \(\chi=\chi_D\) and \[ S=\{p:p\mid2\Delta_1\Delta_2\},\qquad J_S(N)=\prod_{\substack{p\mid N\\p\notin S,\ \chi(p)=-1}} \frac{p-1}{p-2}. \] A target is locally admissible for \(Q_1,Q_2\) if \(N\in\mathscr V_{p,Q_1}+\mathscr V_{p,Q_2}\) for every prime \(p\). For a finite abelian group \(G\), put \[ \eta_*(G)= \begin{cases} +\infty,&2G=0,\\ \displaystyle\min_{\substack{\ell\mid |2G|\\\ell\ {\rm prime}}} \frac{\ell-1}{2\ell},&2G\ne0. \end{cases} \] The mixed local value densities and universal factor are \[ d_{p;1,2}(N)=\int_{\Z_p} \1_{\mathscr V_{p,Q_1}}(a)\1_{\mathscr V_{p,Q_2}}(N-a)\,da \quad(p\in S), \] \begin{equation}\label{eq:form-universal-factor} \mathfrak b_{D,S}=\frac{4L(1,\chi)}\pi \prod_{p\in S}\frac{1-\chi(p)/p}{1-1/p} \prod_{\substack{p\notin S\\\chi(p)=-1}} \left(1-\frac2{p(p-1)}\right). \end{equation} Haar measure gives \(\Z_p\) mass one. \begin{theorem}[Common-field primitive form values]\label{cor:common-field-form-pairs} For the fixed data above, the following assertions hold. \begin{enumerate} \item\label{part:form-classical}\label{thm:genus-binary-invariance} \emph{Classical order bounds and genus comparison.} Uniformly on sufficiently large locally admissible targets, \[ R_{Q_1,Q_2}(N)\asymp R_{\Lambda_{Q_1},\Lambda_{Q_2}}(N) \asymp\frac{NJ_S(N)}{\log N}. \] If \(Q'_i\) belongs to the genus of \(Q_i\), then, for every fixed \(0<\eta<\min_i\eta_*(G_i)\), uniformly on their common sufficiently large locally admissible targets, \[ R_{Q_1,Q_2}(N)=R_{Q'_1,Q'_2}(N) \left(1+O((\log N)^{-\eta})\right). \] If both doubled class groups vanish, the counts agree for every target. Otherwise \(\eta=1/5\) is always permissible. These assertions use Appendix~\ref{sec:older-route} and the class-error sieve, independently of the recent zero-free and weighted-dilation inputs. \item\label{part:form-leading}\label{thm:form-binary-leading} \emph{Explicit leading term.} Under the analytic inputs of Section~\ref{sec:inputs}, uniformly through sufficiently large locally admissible targets, \begin{equation}\label{eq:form-binary-leading} R_{Q_1,Q_2}(N)\sim \mathfrak b_{D,S}\left(\prod_{p\in S}d_{p;1,2}(N)\right) J_S(N)\frac N{\log N}. \end{equation} Each bad-prime density is a computable rational periodic function. The leading coefficient depends only on the respective genera, with no class-number divisor. \end{enumerate} Nonadmissible targets have zero count. Constants and thresholds may depend on the fixed forms and, in the comparison, on \(\eta\). The two conductors may differ; the quadratic field remains fixed. \end{theorem} A saturated numerical value can occur in every class of its compatible genus; the classes do not partition the values. Taking \(f_1=f_2=f\) gives the one-discriminant case, with \(\Delta=f^2D\) and \(J_\Delta=J_S\). Write \(\mathcal F=(Q_1,Q_2)\) for the fixed pair in the proof. \subsection{Proper ideals and the genus map} Fix either order temporarily, write it as \(\mathcal O\), and put \(G=\operatorname{Pic}(\mathcal O)\). The set \(S\) may contain additional primes besides those dividing \(2\operatorname{disc}\mathcal O\). We use proper invertible ideals of the order. If $I$ is an invertible fractional ideal, write $I_p=i_p\mathcal O_p$ and let $n(I)>0$ be its covolume norm. Local determinants give \begin{equation}\label{eq:norm-ideal-identities} v_p(n(I))=v_p(\operatorname N(i_p)),\qquad n(IJ)=n(I)n(J),\qquad I\overline I=n(I)\mathcal O. \end{equation} These statements hold at conductor primes as well. Call an integral ideal primitive if it is not contained in $p\mathcal O$ for any rational prime $p$. Finitely many prescribed locally principal lattices, equal to $\mathcal O_p$ elsewhere, glue to an invertible fractional ideal: place them between $m\mathcal O_p$ and $m^{-1}\mathcal O_p$, glue the residue submodules by the Chinese remainder theorem, and glue their inverses in the same way. The following principal-genus statement is classical for arbitrary orders; compare \cite[Appendix~A, Proposition~A.1]{khayutin}. We give the norm argument because retaining conductor-local ideal types is necessary here. \begin{lemma}\label{lem:norm-genus} Put $U_p=\operatorname N(\mathcal O_p^\times)$. The homomorphism \[ \Theta:G\longrightarrow \prod_{p\in S}\Z_p^\times/U_p,\qquad \Theta_p([I])=n(I)/\operatorname N(i_p)\pmod{U_p} \] is well defined and has kernel $2G$. \end{lemma} \begin{proof} The quotient in $\Theta_p$ is a unit by \eqref{eq:norm-ideal-identities}. A change of local generator changes it by an order-unit norm. Multiplication of $I$ by a principal ideal changes numerator and denominator by the same positive element norm. Every $U_p$ contains the rational unit squares, so the target has exponent two and $2G$ is in the kernel. Outside $S$ every unit is a norm from $\mathcal O_p^\times$. Conversely, if $\Theta([I])=1$, choose units $u_p\in\mathcal O_p^\times$ such that $\operatorname N(i_pu_p)=n(I)$ at every finite prime. Positivity handles the infinite place. The cyclic Hasse norm theorem \cite[Chapter~VIII, Theorem~3.1]{milnecft} supplies $\alpha\in K^\times$ with $\operatorname N(\alpha)=n(I)$. Local Hilbert~90 writes $i_pu_p/\alpha=z_p/\overline z_p$. For a norm-one element $c\ne-1$ one may take $z=1+c$; for $c=-1$, take $z=\sqrt D$. At almost all primes a unit solution exists: in an unramified field normalize its valuation by a rational uniformizer, and in a split algebra use $z=(c_1,1)$. The lattices $z_p\mathcal O_p$ therefore glue to an invertible fractional ideal $J$. They give $I/(\alpha)=J/\overline J$, hence $[I]=2[J]$ by \eqref{eq:norm-ideal-identities}. \end{proof} \subsection{The target ideal class and its local shadow} For either form \(Q(x,y)=ax^2+bxy+cy^2\), use its own order, discriminant and class group, and take \[ \beta=(b+\sqrt\Delta)/2,\qquad A=\Z a+\Z\beta\subset\mathcal O,\qquad c_Q=-[A]\in G. \] Here \(\mathcal O=\Z[\beta]\), \(n(A)=a\), and \(N(ax+\beta y)/a=Q(x,y)\). The ideal is proper invertible: \[ A\overline A=a(a,\beta,\overline\beta,c)\mathcal O=a\mathcal O, \] since \(\beta+\overline\beta=b\) and \(\gcd(a,b,c)=1\). This construction allows \(a\) to share primes with the conductor. It also fixes the class sign in our convention. For the classical normalized ideal-form description compare \cite[Lemma 10.2]{khayutin}. \begin{lemma}\label{lem:form-ideal-shadow} For every positive integer \(n\), \begin{align*} n\in P_Q &\ \Longleftrightarrow\ \text{some primitive integral invertible ideal }I \text{ has }n(I)=n,\ [I]=c_Q,\\ n\in\Lambda_Q &\ \Longleftrightarrow\ \text{some primitive integral invertible ideal }I \text{ has }n(I)=n,\ [I]\in c_Q+2G. \end{align*} \end{lemma} \begin{proof} An element \(\alpha\in A\) gives \(I=(\alpha)A^{-1}\), with \(n(I)=N(\alpha)/a\) and class \(-[A]\). It is integral because \((\alpha)\subset A\). Multiplication by the invertible ideal \(A\) shows \(I\subset p\mathcal O\) if and only if \(\alpha\in pA\). This is exactly the coordinate-primitivity condition. Conversely, if \([I]=-[A]\), then \(IA=(\alpha)\subset A\), and the same identities give a primitive representation of \(n(I)\). Write \(A_p=z_p\mathcal O_p\). The norm identities in \eqref{eq:norm-ideal-identities} give the units \[ \lambda_p=N(z_p)/a\in\Z_p^\times,\qquad \mathscr V_{p,Q}=\lambda_p\mathscr V_p, \] where \(\mathscr V_p=\{N(w):w\in\mathcal O_p\setminus p\mathcal O_p\}\) is the primitive norm-value set for \(\mathcal O_p\). The change from an integral basis of \(A\) to the local generator is in \(\operatorname{GL}_2(\Z_p)\). Moreover \(\Theta_p(c_Q)=\lambda_p\pmod{U_p}\), with \(\Theta\) and \(U_p\) from Lemma~\ref{lem:norm-genus}. If \(n\in\Lambda_Q\), choose primitive local elements \(\alpha_p=z_pw_p\in A_p\) with \(N(\alpha_p)=an\). The ideals \(w_p\mathcal O_p\) are primitive integral invertible, and equal \(\mathcal O_p\) outside finitely many primes. Glue them as in Lemma~\ref{lem:norm-genus}. Their norm valuations are \(v_p(n)\), so the resulting ideal \(I\) has norm \(n\), and \[ \Theta_p([I])=n/N(w_p)=\lambda_p\pmod{U_p}. \] The kernel of \(\Theta\) is \(2G\), proving the coset assertion. Conversely, for such an \(I\), write \(I_p=i_p\mathcal O_p\). Equality of the \(\Theta\)-classes supplies a unit \(u_p\) with \(N(u_p)=n/(\lambda_pN(i_p))\). Then \(\alpha_p=z_pi_pu_p\) is primitive in \(A_p\) and has norm \(an\). Outside \(S\) the order-unit norm group is all of \(\Z_p^\times\), so the same conclusion holds there. This proves the local-shadow equivalence at conductor primes as well. \end{proof} The coset condition is existential. A conductor-divisible value need not determine a unique genus. For example, at discriminant \(-32\), \(x^2+8y^2\) and \(3x^2+2xy+3y^2\) have primitive odd values \(1\pmod8\) and \(3\pmod8\), respectively, and hence different genera. Both nevertheless primitively represent \(24\), at \((4,1)\) and \((3,-1)\). \subsection{Orientation saturation and maximal subgroups} For either form \(Q\), work in its own group \(G\), with target class \(c_Q=-[A]\), and write $h_{\mathcal O}=|G|$ and $B=2G$. For every inverse pair $\{g,-g\}\subset G$ with $2g\ne0$, let $\Omega_g$ be the rational split primes outside $S$ whose two prime ideal classes are $g,-g$, and put \[ m_g=\operatorname{ord}(2g)-1,\qquad k_g(n)=\#\{p\in\Omega_g:v_p(n)=1\}. \] These prime families are disjoint. Enlarging \(S\) deletes only finitely many primes. The fixed-discriminant prime-class estimate \cite[Lemma~2, equation~(5.2)]{bourgainfuchs} gives \begin{equation}\label{eq:norm-class-mertens} \sum_{\substack{p\le y\\p\in\Omega_g}}\frac1p =\frac1h_{\mathcal O}\log\log y+O_{\mathcal O,S}(1). \end{equation} The coefficient is $1/h_{\mathcal O}$ for a nonambiguous inverse pair. The source allows nonfundamental negative form discriminants. For a fixed order its prime-counting error is $O_{\mathcal O}(x\exp(-c\sqrt{\log x}))$; partial summation gives the bounded remainder in \eqref{eq:norm-class-mertens}. Call the pair $\{g,-g\}$ saturated for $n$ if $k_g(n)\ge m_g$, and set \[ B_n=\langle2g:\{g,-g\}\text{ is saturated for }n\rangle\le B. \] If $n\in\Lambda_Q$ and $B_n=B$, then $n\in P_Q$. To see this, start with a primitive ideal from Lemma~\ref{lem:form-ideal-shadow}, whose class belongs to $c_Q+B$. At a good split prime, primitivity puts the entire prime power in one conjugate prime ideal. From each saturated family reserve $m_g$ exponent-one factors, initially orienting each as $g$. This changes the ideal class only by an element of $B$, so its class still belongs to $c_Q+B$. Reversing $j$ reserved orientations changes its class by $-2jg$, for $0\le j\le m_g$; these changes fill $\langle2g\rangle$. Families have disjoint prime support, so their choices are independent and their sums fill $B_n=B$. They therefore cancel its difference from \(c_Q\). The resulting ideal is primitive, has class \(c_Q\), and has the same numerical norm and every conductor-local ideal unchanged. Hence $n\in P_Q$ by Lemma~\ref{lem:form-ideal-shadow}. Suppose $B\ne0$, and let $\mathcal H$ be its finite set of maximal proper subgroups. For $M\in\mathcal H$ define \[ \begin{gathered} \ell_M=[B:M],\qquad \Omega_M=\bigcup_{\{g,-g\}:\,2g\notin M}\Omega_g,\\ K_M=\sum_{\{g,-g\}:\,2g\notin M}(m_g-1),\qquad k_M(n)=\#\{p\in\Omega_M:v_p(n)=1\}. \end{gathered} \] Each inverse pair is counted once. Since $B$ is finite abelian, $B/M$ is cyclic of prime order $\ell_M$. If $n\in\Lambda_Q\setminus P_Q$, then $B_n$ is proper and lies in some $M\in\mathcal H$. Every pair with $2g\notin M$ is consequently deficient, giving \begin{equation}\label{eq:norm-deficiency} \Lambda_Q\setminus P_Q\subset \bigcup_{M\in\mathcal H}\{n:k_M(n)\le K_M\}. \end{equation} If $2G=0$, Lemma~\ref{lem:form-ideal-shadow} already gives $P_Q=\Lambda_Q$. The doubling map $G\to B$ has equal fibres. Precisely $h_{\mathcal O}(\ell_M-1)/\ell_M$ classes satisfy $2g\notin M$. None is ambiguous, since $2g\ne0$, so they form $h_{\mathcal O}(\ell_M-1)/(2\ell_M)$ inverse pairs. Summing \eqref{eq:norm-class-mertens} over these disjoint families proves \begin{equation}\label{eq:norm-maximal-mertens} \sum_{\substack{p\le y\\p\in\Omega_M}}\frac1p =d_M\log\log y+O_{\mathcal O,S}(1),\qquad d_M=\frac{\ell_M-1}{2\ell_M}\ge\frac14. \end{equation} Every prime divisor of $|B|$ occurs as the index of a maximal subgroup: in the corresponding primary component, reduce a cyclic factor modulo that prime. Thus $\min_{M\in\mathcal H}d_M=\eta_*(G)$. The proper-subgroup mechanism is classical in the exceptional-value argument of Bourgain--Fuchs \cite[Theorem 1, alternative (4.2), and Section 2, part (5)]{bourgainfuchs}. Here its use in \eqref{eq:norm-deficiency} makes an entire positive-density collection deficient on the binary line. For $G\simeq\Z/8\Z$ or $\Z/2\Z\times\Z/4\Z$ the threshold is $1/4$, while for $G\simeq(\Z/3\Z)^2$ it is $1/3$. \begin{proposition}[The class error]\label{prop:norm-class-error} In the common-field setting above, for every fixed \(0<\eta<\min_i\eta_*(G_i)\), uniformly for \(N\ge3\), \begin{equation}\label{eq:form-class-error}\csname ltx@label\endcsname{eq:norm-class-error} 0\le R_{\Lambda_{Q_1},\Lambda_{Q_2}}(N)-R_{Q_1,Q_2}(N) \ll_{\mathcal F,\eta} \frac{NJ_S(N)}{(\log N)^{1+\eta}}. \end{equation} If \(2G_i=0\), then \(P_{Q_i}=\Lambda_{Q_i}\), and that summand contributes no class error. If both doubled groups vanish the displayed difference is zero. In every case the exponent \(\eta=1/5\) is permissible. Only this saving exponent is independent of the orders; the implied constant remains dependent on the fixed forms. The proposition is classical and uses neither recent analytic input. \end{proposition} \subsection{A sieve for the class error} We prove Proposition~\ref{prop:norm-class-error} directly on the line $a+(N-a)=N$. First charge a bad pair to its first summand \(a\in\Lambda_{Q_1}\setminus P_{Q_1}\), and use the maximal-subgroup families in \(G_1\). If \(2G_1=0\), there is no such error. Otherwise fix \(M\in\mathcal H\) in this group. At every good inert prime $p$, both values must be units, so the forbidden residues of $a$ modulo $p$ are $0,N$, with $\nu_N(p)=2-\mathbf1_{p\mid N}$ roots. At a marked prime $p\in\Omega_M$, forbid exactly $v_p(a)=1$: this removes the $p-1$ residues $p,2p,\ldots,(p-1)p$ modulo $p^2$. Its forbidden density is \[ f_*(p)=\frac{p-1}{p^2}. \] Higher prime powers remain permitted, including when $p\mid N$. Here are the needed uniform sieve details. Given any mark set $\mathcal M\subset\Omega_M\cap[1,y]$, use modulus $p$ at inert primes and $p^2$ at marked primes. All sieve indices below are squarefree and supported on the chosen inert and marked primes at most $y$. For such an index $d$, let $\ell(d)$ be the product of these moduli, $\nu(d)$ the product of the root counts, and $f(d)=\nu(d)/\ell(d)$. Exact interval counting by the Chinese remainder theorem gives $Xf(d)+O(\nu(d))$ on an interval of length $X$. In all cases $\nu(d)\le d$ and $01/5$. This argument uses only the existing fixed-order prime-class estimate, CRT and the finite upper sieve, independently of the recent zero-free and graph theorems. The use of \(t<1\) explains the strict range \(\eta<\min_i\eta_*(G_i)\); no endpoint or optimality assertion is made. \subsection{Two local filters on a common split core}\label{sec:form-local-assembly} The remaining calculation concerns the two local shadows. It is shared by all pairs above and by the principal specialization below. First the local sets have a finite description, even when a split conductor prime permits unbounded valuations. Put \[ e_p=1+2\max_i v_p(\Delta_i)\qquad(p\in S). \] At a primitive vector \(v\), the identity \[ \operatorname{adj}(\operatorname{Hess}Q_i)\nabla Q_i(v) =-\Delta_i v \] shows that some partial derivative has valuation \(s\le v_p(\Delta_i)\). For any target congruent to \(Q_i(v)\) modulo \(p^{e_p}\), strong one-variable Hensel lifting in that coordinate applies because \(e_p>2s\). The lifted vector stays primitive. Thus each \(\mathscr V_{p,Q_i}\) is a union of full cosets modulo \(p^{e_p}\), found by enumerating primitive vectors in that quotient, and \begin{equation}\label{eq:norm-local-density-finite} d_{p;1,2}(N)=\frac1{p^{e_p}} \#\left\{a\bmod p^{e_p}: \begin{array}{l} a\in\mathscr V_{p,Q_1}\bmod p^{e_p},\\ N-a\in\mathscr V_{p,Q_2}\bmod p^{e_p} \end{array}\right\}. \end{equation} In particular these densities are rational periodic functions. Outside \(S\), the primitive value set of either form is \(\Z_p\) at split primes and \(\Z_p^\times\) at inert primes. These primes are odd, so every target is a sum of two permitted local values. Local admissibility is therefore the finite criterion \(\prod_{p\in S}d_{p;1,2}(N)>0\), and this product has a fixed positive minimum on admissible targets. The local genus biases must still be retained. Set \[ L=8\prod_{\substack{p\in S\\p\text{ odd}}}p,\qquad \alpha=\varphi(L)/L,\quad H=\ker(\chi\bmod L),\quad \mathcal C=\langle p:p\notin S,\ \chi(p)=1\rangle. \] For each \(S\)-supported \(r\), put \[ A_{r,i}=\{u\in(\Z/L\Z)^\times: ru\in\mathscr V_{p,Q_i}\text{ for all }p\in S\}. \] These sets are well defined and are unions of unit-square states: equal units modulo an odd \(p\), or modulo eight at two, differ by a local unit square; multiplication of a primitive vector by a rational unit scales its value by that square. Crucially \(A_{r,i}\subset H\), although individual local genus characters can be nonprincipal. To verify this, use an ideal \(A_i\) as above, with norm \(a_i\) and local multipliers \(\lambda_{i,p}\). At every prime outside \(S\), an ideal norm has even valuation at an inert prime and has no field-norm obstruction at a split prime. Positivity and Hilbert reciprocity therefore give \[ \prod_{p\in S}(\lambda_{i,p},D)_p =\prod_{p\in S}(a_i,D)_p=1. \] Here the numerator in \(\lambda_{i,p}=N(z_{i,p})/a_i\) is a field norm. If \(u\in A_{r,i}\), then \((ru,D)_p=(\lambda_{i,p},D)_p\) at every bad prime. The positive \(S\)-supported number \(r\) has product of bad symbols one. For an integer unit representative \(u\), reciprocity identifies \(\prod_{p\in S}(u,D)_p\) with \(\chi(u)\). Multiplication of the displayed local relations gives \(\chi(u)=1\). This does not identify \(A_{r,i}\) with the principal-form filter. Consequently the exact disjoint decomposition is \begin{equation}\label{eq:form-prefix-decomposition}\csname ltx@label\endcsname{eq:norm-prefix-decomposition} \Lambda_{Q_i}=\coprod_{r\in\langle S\rangle} r\{x\in\mathcal C:x\bmod L\in A_{r,i}\}. \end{equation} Let $c_{\mathcal C}$ denote the Gamma-normalized marginal constant of this core. Its Euler series satisfies \[ \left(\sum_{n\in\mathcal C}n^{-s}\right)^2 =\zeta(s)L(s,\chi) \prod_{p\in S}(1-p^{-s})(1-\chi(p)p^{-s}) \prod_{\substack{p\notin S\\\chi(p)=-1}}(1-p^{-2s}), \] so \begin{equation}\label{eq:norm-core-constant} c_{\mathcal C}^2=\frac{L(1,\chi)}\pi \prod_{p\in S}(1-p^{-1})(1-\chi(p)p^{-1}) \prod_{\substack{p\notin S\\\chi(p)=-1}}(1-p^{-2}). \end{equation} Also put \[ C_{\rm in}=\prod_{\substack{p\notin S\\\chi(p)=-1}} \left(1-\frac1{(p-1)^2}\right). \] \subsection{Finite profiles, the classical bounds, and genus comparison} For fixed prefixes \(r,s\in\langle S\rangle\), put \(g=(r,s)\). When \(g\mid N\), write \(A=r/g\), \(B=s/g\), \(M=N/g\), and set \[ \Gamma_{r,s;1,2}(N)=\frac L{|H|^2} \#\{(u,v)\in A_{r,1}\times A_{s,2}:Au+Bv=M\bmod L\}; \] otherwise set it to zero. The original two filters are retained after dividing by \(g\). A finite collection of profiles covers every locally admissible target. Put \(Z=\prod_{p\in S}p^{e_p+1}\). Given local summands \(a+b=N\), replace them by \(a+p^{e_p}t,b-p^{e_p}t\). They remain in their full value cosets. A valuation below \(e_p\) stays fixed; otherwise at most one residue \(t\bmod p^2\) makes that summand divisible by \(p^{e_p+2}\). Since \(p^2>2\), both forbidden residues can be avoided. The summands can therefore be chosen with valuations at most \(e_p+1\). Their valuations determine \(r,s\mid Z\), and CRT assembles their unit quotients into a root counted by \(\Gamma_{r,s;1,2}(N)\). Conversely, any such root lifts to an exact linear solution at each bad prime: one of \(A,B\) is a unit, and fixing the other unit residue preserves the two unit-square states. The local definitions of \(A_{r,1},A_{s,2}\) therefore remain valid. Outside \(S\) there is already no additive obstruction. Hence \[ N\text{ locally admissible}\quad\Longleftrightarrow\quad \Gamma_{r,s;1,2}(N)>0\text{ for some }r,s\mid Z. \] \begin{proof}[Proof of Theorem~\ref{cor:common-field-form-pairs}, part~\ref{part:form-classical}] Lemma~\ref{lem:older-quadratic-filter} allows two distinct filters \(U,V_0\subset H\). In a positive fixed profile it supplies \(\gg_{\mathcal F}MJ_S(M)/\log M\) core pairs on \(Ax+By=M\). Multiplication by \(r,s\) is injective, and \(J_S(M)=J_S(N)\) because \(g\) is supported on \(S\). Only finitely many profiles and root patterns occur above, so the minimum of the positive lower constants and maximum of their thresholds give \[ R_{\Lambda_{Q_1},\Lambda_{Q_2}}(N) \gg_{\mathcal F}NJ_S(N)/\log N \] uniformly on sufficiently large admissible targets. Sieving the original line only at good inert primes gives the matching upper bound by the unmarked case of \eqref{eq:norm-marked-sieve}. Proposition~\ref{prop:norm-class-error} transfers the lower bound to \(R_{Q_1,Q_2}\). If \(Q'_i\) is in the genus of \(Q_i\) for each \(i\), their local value sets coincide exactly, as do the two mixed local shadows. Subtract their two class-error estimates and divide by the classical positive lower bound. This proves part~\ref{part:form-classical} of Theorem~\ref{cor:common-field-form-pairs}. If both doubled groups vanish the individual global and local value sets are equal, proving the exact assertion for every target. Nonadmissible targets have zero count. This argument uses the older quadratic filter estimate, not Theorem~\ref{thm:general}. \end{proof} \subsection{Prefix summation and Haar normalization} Under the inputs of Section~\ref{sec:inputs}, apply Theorem~\ref{thm:general} with the coprime fixed \(L\)-supported coefficients \(A,B\) in each profile. Write \(R_{r,s;1,2}(N)\) for its contribution in \eqref{eq:form-prefix-decomposition}. Since \(J_S(N/g)=J_S(N)\), its contribution divided by \(NJ_S(N)/\log N\) is \begin{equation}\label{eq:norm-prefix-main} c_{\mathcal C}^2 C_{\rm in}\Gamma_{r,s;1,2}(N)\frac g{rs}+o(1). \end{equation} The original filters \(A_{r,1},A_{s,2}\) remain in this formula. To sum the fixed-profile asymptotics, parameterize \(rx+sy=N\). Its positive interval has length at most \(Ng/(rs)\). At any good inert prime the two coefficients in this parameterization are units, and its two forbidden roots merge exactly when the prime divides \(N\). Discarding both residue filters and applying the unmarked sieve \eqref{eq:norm-marked-sieve} with \(z=N^{1/10}\) gives \begin{equation}\label{eq:norm-prefix-tail} R_{r,s;1,2}(N)\ll_{\mathcal F} \frac{NgJ_S(N)}{rs\log N}+O(N^{9/20}), \end{equation} uniformly in \(r,s\). The interval error also covers intervals shorter than the sieve level. Only \(O_{\mathcal F}((1+\log N)^{2|S|})\) prefix pairs can occur, while \[ \sum_{r,s\in\langle S\rangle}\frac{(r,s)}{rs} =\prod_{p\in S}\sum_{i,j\ge0}p^{-\max(i,j)} =\prod_{p\in S}\frac{1+p^{-1}}{(1-p^{-1})^2}<\infty. \] The total interval error is negligible relative to \(NJ_S(N)/\log N\); the remaining tail is uniformly small as a fixed prefix bound tends to infinity. This justifies a finite-prefix limit followed by a summable-tail limit, without assuming a binary asymptotic uniform in growing coefficients. Partition the local Haar line \(a+b=N\) by the exact valuation prefixes \(r,s\). At a prime with \(v_p(r)=i\), \(v_p(s)=j\), and \(v_p(L)=k\), each compatible unit-pair root modulo \(p^k\) has measure \(p^{-\max(i,j)-k}\). For \(i\le j\), prescribing the second summand modulo \(p^{j+k}\) fixes the first summand's required unit residue; the case \(j\le i\) is symmetric. Thus each global root has measure \(g/(Lrs)\), and the full stratum has measure \[ \frac g{Lrs} \#\{(u,v)\in A_{r,1}\times A_{s,2}:Au+Bv=M\bmod L\} =\frac{\alpha^2}{4}\Gamma_{r,s;1,2}(N)\frac g{rs}, \] because \(|H|=\varphi(L)/2\). Zero values have Haar measure zero. Summing the disjoint strata gives the single normalization identity \begin{equation}\label{eq:form-haar-strata}\csname ltx@label\endcsname{eq:norm-haar-strata} \sum_{r,s}\Gamma_{r,s;1,2}(N)\frac g{rs} =\frac4{\alpha^2}\prod_{p\in S}d_{p;1,2}(N). \end{equation} Combining this identity with \eqref{eq:norm-prefix-main} and the summable tail proves \[ R_{\Lambda_{Q_1},\Lambda_{Q_2}}(N)\sim \frac{4c_{\mathcal C}^2}{\alpha^2}C_{\rm in} \left(\prod_{p\in S}d_{p;1,2}(N)\right) J_S(N)\frac N{\log N}. \] The coefficient is explicit. In \eqref{eq:norm-core-constant}, divide the bad-prime factors by \(\alpha^2=\prod_{p\in S}(1-p^{-1})^2\), and use \[ (1-p^{-2})\left(1-\frac1{(p-1)^2}\right) =1-\frac2{p(p-1)} \] at every good inert prime. The result is precisely \(\mathfrak b_{D,S}\) in \eqref{eq:form-universal-factor}. The finite positive minimum of the local density product and \eqref{eq:form-class-error} transfer the asymptotic to the actual values. This proves part~\ref{part:form-leading} of Theorem~\ref{cor:common-field-form-pairs}.