\section{Disproof of the unrestricted fixed-section channel bound} \label{uc:section} The construction in this section exploits the distinction between power crossing a chosen square and power delivered to a receiver on its far side. Near an edge, a circulating field can cross into the square and return outside it. The positive contribution inside the square counts as passage; the return flow contributes to absolute bypass. The definition permits this provided both are within their prescribed thresholds. Selecting the complete outgoing scattering map also counts reflected radiation in the output power. The construction makes this picture work simultaneously for any prescribed finite number of coherent input directions. Small dielectric spheres near paired edges produce localized circulation. A distant radial stack amplifies the incident fields that drive them, compensating for their shrinking size. The stack and spheres are ordinary finite dielectrics; their size and number of layers may depend on the requested dimension. A uniform estimate for all superpositions, including interference terms and the whole measurement plane, connects this picture to the actual Maxwell fields. The unrestricted target quantifies over all fixed passage and bypass thresholds in Definition~\ref{pr:count}. The following counterexample uses a permitted range of those thresholds. It does not address the different question obtained by requiring $\delta\leq\eta$, or by restricting the selected outputs to a specified receiving hemisphere. \begin{theorem}[Arbitrarily large counted subspaces at fixed thresholds] \label{uc:counterexample} Fix $k_0>0$, $\eps_+>1$, $L>0$, $0<\tau<1$, and $0<\eta<\delta$. For every positive integer $N$ there is a finite, piecewise smooth, real scalar dielectric with $1\leq\eps_r\leq\eps_+$, vacuum exterior, and no interior sources, whose physical incoming space contains an $N$-dimensional counted subspace through the same square $D=(-L/2,L/2)^2\times\{0\}$. The selected outgoing map is its complete scattering map. In particular, every vector in this subspace has unit power throughput relative to its incident power, passage at least $\eta$ times that power, and absolute bypass at most $\delta$ times that power. All plane fluxes are integrable. Consequently the unrestricted finiteness assertion in Conjecture~\ref{pr:channel-target}, and hence its stronger area bound, is false. The frequency, material interval, square and thresholds are common to the whole sequence of devices. \end{theorem} The strict inequality $\eta<\delta$ leaves room to choose an intermediate power $p_0$: in the limiting construction the passage and absolute bypass both approach $p_0$ for every unit input in the counted subspace. The proof then chooses a finite device close enough to this limit to satisfy both inequalities. Since the complete lossless scattering map preserves incident power, the throughput condition holds for every $\tau<1$. This does not require that power leaving on a particular side be collected by a receiver. The order of limits matters. For each requested dimension, first choose the independent incident data; then increase the stack gain while shrinking the spheres; finally take the stack radius large enough to approximate the limiting Green kernel for that particular choice. No single device or uniform radius is claimed to work for all dimensions at once. \begin{proof} Choose once and for all an index $1\rho>0$. Choose $p_0\in(\eta,\delta)$ and then $t>0$ so that $\alpha t^2/(2\pi h)=p_0$, where $\alpha=4\pi\rho^3(\eps_p-1)/(\eps_p+2)>0$. These choices do not depend on $N$. Fix $N$. Lemmas~\ref{ct:jets}--\ref{ct:gram} supply $N$ distinct right-edge points and their left-edge partners, a finite band of regular electric-wall TE interpolants $v_j$, and $\lambda>0$ such that $g_j=v_j/\sqrt\lambda$ are orthonormal physical incoming data with the prescribed independent jets. Let \[ B_M=q^M,\qquad a_M=t\sqrt\lambda/B_M,\qquad \ell_0=k_0^{-1}. \] Thus $a_M\to0$ and the actual common core gain $B_M$ produces the jet multiplier $A_M=B_M/\sqrt\lambda=t/a_M$. Place the $2N$ radius-$\rho a_M\ell_0$ spheres at height $h a_M\ell_0$ above the chosen edge points. For large $M$ they are disjoint, lie strictly above the plane, and remain inside a fixed ball. Their arrangement is invariant under rotation by $\pi$ about the $z$ axis. For each such $M$, Theorem~\ref{cl:remainder} proves invertibility of the actual inclusion volume equation for the explicit limiting background kernel with gain $B_M$. It gives, uniformly for $c\in\mathbb C^N$ with $\|c\|=1$, \[ \|S_{z,M}^{\infty}(c)-F_{a_M}(c)\|_{L^1(\mathbb R^2)} \leq C_N a_M. \] The constant is independent of $M$ once $N$ and its interpolants are fixed. This estimate includes the self-consistent polarization, all source angular orders, all coherent cross terms, and the entire measurement plane. Now hold $M$ fixed. Take a common radial stack of $M$ high-low quarter-wave pairs, with indices $q$ and one, starting at dimensionless radius $R=2\pi n$. The physical radius is $R/k_0$. Theorem~\ref{rg:physical-limit} proves both convergence of its source Green kernel to that used above and convergence in whole-plane $L^1$ of the resulting \emph{actual perturbed-device} fluxes, uniformly on this finite input space. Inverse stability supplies the actual dielectric polarizations, not arbitrary imposed dipole sources. Choose a sufficiently large finite integer $n=n(M,N)$ so that \[ \sup_{\|c\|=1} \|S_{z,M}^{\rm phys}(c)-S_{z,M}^{\infty}(c)\|_1\eta$; it does not refute a reformulated claim with smaller bypass or additional output restrictions. In particular the resource-controlled theorems requiring $\delta<\eta\tau$ retain their stated scope. The total support, common core amplification, interface count and inverse minimum feature size grow along the construction. None is a bounded resource in Conjecture~\ref{pr:channel-target}. The proof is analytic. Its complete physical construction and the same-device passivity identity are formalized in the endpoint listed in Section~\ref{ver:lean}; no finite-element simulation is claimed. \end{remark}