\section{Analytic inputs and the comparison framework}\label{sec:inputs} We use Walsh's comparison architecture \cite[Sections 2--7]{walsh}, the zero-free input \cite[Theorem 3.1]{zerofree}, and the weighted-dilation transfer \cite[Section 3]{weightedgraphs}. The operator normalizations and parameter order are stated explicitly. \subsection{The zero-free input} \begin{proposition}[Imported zero-free theorem]\label{prop:zf} Every Dirichlet \(L\)-function, including \(\zeta\), is nonzero in \[ \Re s>11/12. \] The pole at \(s=1\) of a principal character is allowed. Both conductor and height are unrestricted. \end{proposition} This is \cite[Theorem 3.1, printed p.~10]{zerofree}, restricted to Dirichlet characters and the open half-plane. We also use the usual functional equation, Euler product and holomorphic logarithm on zero-free rectangles. \subsection{Prime groups and the two operators} For this subsection only, write \[ \Lambda=\log x,\qquad T=\log\Lambda. \] Fix constants \[ 0<\alpha<\beta<\gamma<\delta<0.47,\quad c_0,C_0,v_-,v_+>0,\quad 0<\vartheta<1,\quad \ell\ge1. \] There are disjoint collections \(\mathcal G_s,\mathcal G_b\) of pairwise disjoint finite prime sets \(P_g\), with \[ c_0T\le|\mathcal G_s|,|\mathcal G_b|\le C_0T. \] They satisfy \begin{equation} \begin{split} e^{\Lambda^\alpha}\le p\le e^{\Lambda^\beta} &\quad(g\in\mathcal G_s,\ p\in P_g),\\ e^{\Lambda^\gamma}\le p\le e^{\Lambda^\delta} &\quad(g\in\mathcal G_b,\ p\in P_g),\\ v_-\le V_g:=\sum_{p\in P_g}\frac1p\le v_+, &\qquad \mu_g(p)=\frac1{V_gp}. \end{split} \label{eq:inputbanks} \end{equation} The sets need not be intervals. This freedom is needed when taking identity residue classes and deleting primes dividing the target. Fix a sufficiently large integer \(J\), and put \(M=J+\ell\). A pattern \(\nu\) specifies subsets \(C_g\subseteq\{1,\ldots,J\}\), with \[ C_g=\varnothing\quad(g\in\mathcal G_s),\qquad |C_g|\le m\quad(g\in\mathcal G_b),\qquad t_g=\ell+|C_g|. \] The slots in \(\{1,\ldots,J\}\setminus C_g\) are shared between the two endpoint lists. The remaining \(t_g\) slots at each endpoint are unshared. Form \(D\) from the shared labels and, in the \(C_g\)-slots, fresh independent draws with laws \(\mu_g\). Thus \(D\) has \(J\) prime factors from each group, counted with multiplicity. Write \(D=D_sD_b\) according to group size. Each pattern has a complex coefficient \(K_\nu\). It may depend on the ordered big-group source labels, target labels, and auxiliary labels used in \(D_b\), and on no small-group labels. Assume \begin{equation} \sum_\nu\sup |K_\nu|\le\Lambda^{C_1}, \label{eq:inputmass} \end{equation} where \(m,C_1\) are fixed independently of \(J\). All operators below average source and target slot permutations independently in each applicable group. These averages are orthogonal projections. The \emph{ideal operator} acts on \[ \mathcal H_{\mathrm{id}}= L^2\!\left(\prod_{g\in\mathcal G_b}P_g^M, \bigotimes_{g\in\mathcal G_b}\mu_g^{\otimes M}\right). \] Numerical repetitions are permitted. Its row operation retains shared labels, samples the unshared target labels and auxiliary labels independently, and multiplies by \[ K_\nu D_b^{\,i\zeta}e(\Theta D_b), \qquad e(t)=\exp(2\pi it). \] Sum over patterns and compose with the permutation projections on both sides; denote the result by \(\mathcal T(\Theta)\). A \emph{physical state} consists of an integer \(n\) and an ordered \(M\)-tuple in every group, with distinct labels within each group, all dividing \(n\). Give every such state at \(n\) mass \(\prod_gV_g^{-M}\), and use counting measure on the integer position. This defines \(\mathcal H_{\mathrm{ph}}\). Put \[ \omega_g(n)=\#\{p\in P_g:p\mid n\},\quad \omega(n)=\sum_g\omega_g(n),\quad s=|\mathcal G_s|+|\mathcal G_b|. \] For a fixed integer \(k\) with \(0<|k|\le\Lambda^{C_2}\), the \emph{physical operator} \(A\) moves from \(n\) to \(n'=n+kD\). It sums over the physical target lists with the prescribed shared labels, using row normalization \(\prod_gV_g^{-t_g}\); auxiliary labels retain their independent probability laws. Every auxiliary label is forbidden from both endpoint lists, although auxiliary labels may equal each other. Multiply the row by \begin{equation} K_\nu D^{\,i\zeta} \vartheta^{(\omega(n)-Ms)/2} \vartheta^{(\omega(n')-Ms)/2}, \label{eq:inputphysical} \end{equation} sum over patterns and apply all endpoint permutation projections. The adjoint reverses the displacement and conjugates the coefficient with the endpoint roles interchanged. The joint endpoint normalization in each group is \(V_g^{-M-t_g}\) in either direction. \begin{proposition}[Imported absolute bounds]\label{prop:graphabsolute} Replace every elementary transition coefficient of \(A\) by its absolute value before summing patterns and averaging permutations. The resulting operator and its adjoint have row sums at most \(\Lambda^{C_{\mathrm{abs}}}\). The exponent may depend on \(m,C_1\) and the fixed group data, but is independent of \(J\). The same assertion holds with integer position replaced by independent uniform residues at all group primes. \end{proposition} This is \cite[Lemma 3.4]{weightedgraphs}. The independence from \(J\) is part of the assertion. To see its source, once the shared labels are fixed, the new target slots in a group cost at most \[ \sup_{z\ge0}V_g^{-t_g}(z+t_g)^{t_g}\vartheta^{z/2}. \] Here \(t_g\le\ell+m\). This is a fixed constant per group and its product over \(O(T)\) groups is a fixed power of \(\Lambda\). Reversing the joint endpoint measure gives the column bound. \subsection{Root moments, residue replacement, and endpoints} \begin{proposition}[Imported local transference]\label{prop:rootmoment} For every fixed \(E>0\), there is \(E_{\mathrm{id}}>0\), depending only on \(E\) and the fixed group data, such that the following holds. Suppose \begin{equation} \sup_{\Theta\in\mathbb R} \|\mathcal T(\Theta)\|_{2\to2} \le\Lambda^{-E_{\mathrm{id}}}. \label{eq:inputideal} \end{equation} After \(m,C_1\) have been fixed, take a sufficiently large fixed \(J\). Put \(R=\lceil\Lambda^{0.52}\rceil\). For every fixed \(\eta>0\) and \(I=[X,2X)\), with \(x^\eta\le X\le x^{1/\eta}\), \begin{equation} \frac1{|I\cap\Z|}\sum_{n\in I\cap\Z} \langle u_n,(AA^*)^Ru_n\rangle \le\Lambda^{-2RE}. \label{eq:inputroot} \end{equation} Here \(u_n\) is one on every physical mark list at position \(n\) and zero elsewhere. The threshold may depend on all the fixed parameters, including \(J,C_2,\eta\). No bound on \(\zeta\) is required beyond \eqref{eq:inputideal}. \end{proposition} This is \cite[Theorem 3.5]{weightedgraphs}. It imposes return of the integer position. The final mark list need not equal the initial list, and \(u_n\) is not normalized. Substituting a trace with identical initial and final mark lists would change the statement. The arithmetic replacement inside that theorem has the following precise form, which also explains its scale limitation. For a fixed compatible path with \(r=2R\) edges, let \(h_i\) be the integer offsets and \(\mathcal A_i\) its active prime set at visit \(i\). Set \(\vartheta_i=\vartheta\) at internal visits and \(\vartheta_0=\vartheta_r=\sqrt{\vartheta}\). The position-dependent factor is \begin{equation} W(n)=\prod_{i=0}^{r} \left(\prod_{p\in\mathcal A_i}1_{p\mid n+h_i}\right) \vartheta_i^{\,\#\{p\notin\mathcal A_i: \ p\text{ is a group prime},\ p\mid n+h_i\}}. \label{eq:inputpath} \end{equation} \begin{proposition}[Imported residue replacement]\label{prop:graphcrt} For every fixed \(B>0\), the average of \eqref{eq:inputpath} on \(I\cap\Z\) equals its expectation with independent uniform residues at all group primes, multiplied by \(1+O(\exp(-BrT))\). An incompatible collection of active congruences has zero weight in both models. \end{proposition} This is \cite[Lemma 3.6]{weightedgraphs}. For clarity, its relative form retains all deterministic damping belonging to active primes. The product \(Q_{\mathrm{act}}\) of active primes satisfies \[ \log Q_{\mathrm{act}} =O_J(rT\Lambda^\delta)=o(\Lambda). \] After those factors are extracted, a Bonferroni expansion of degree \(O_B(rT)\) controls the remaining damping. Its independent expectation is at least \(\exp(-O(rT))\); its total interval counting error is \(\exp(-\eta\Lambda+o(\Lambda))\). The inequality \(0.52+\delta<1\) makes that error negligible even after division by the active congruence density. Together with Proposition~\ref{prop:graphabsolute}, this permits summation over paths with absolute coefficients. These observations describe the imported replacement and do not assert replacement on intervals shorter than its stated range. \begin{corollary}[Imported endpoint consequence]\label{cor:graphendpoint} Under Proposition~\ref{prop:rootmoment}, let \(f\in\mathcal H_{\mathrm{ph}}\) be supported on positions \(I=[X,2X)\), constant on the chosen mark list at each position, with value \(f_n\). If \[ \sup_{n\in I}|f_n|\le\exp(O(\sqrt{\Lambda})),\qquad \|f\|,\|g\|\le X^{1/2}\Lambda^{C_3}, \] then \begin{equation} |\langle f,Ag\rangle| \ll X\Lambda^{-E+2C_3}. \label{eq:inputendpoint} \end{equation} Only \(f\) must be constant on mark lists. \end{corollary} This is \cite[Corollary 3.11]{weightedgraphs}. Its proof partitions \(I\) into intervals on the displacement scale \(\exp(O_J(T\Lambda^\delta))\), uses spectral H\"older for \(AA^*\), and bounds the positive matrix \((\langle u_n,(AA^*)^Ru_m\rangle)\) by its trace. It makes no equal-fiber-mass assumption. \subsection{The application interfaces} Three steps remain: construct the fractional-density models retaining every surviving character; prove the progression-filter ideal comparison with its big-label coefficient bound \eqref{eq:inputmass} fixed before \(J\); and transfer to the growing reflected target with uniform endpoint and localization bounds. The root-moment theorem supplies none of these steps on its own. Sections~\ref{sec:graph}--\ref{sec:assembly} prove the comparison; the applications then check its arithmetic hypotheses.