\section{Verification, scope, and separate verdicts} \label{ver:section} \subsection{What was checked} The hotspot, channel, and counterexample tracks used disjoint source-file ownership. A different research agent reviewed each track's key arguments, and the coordinator checked their equations and scope. The review found and corrected the direction of the bound on the real parts of the outgoing Dirichlet-to-Neumann multipliers, and made explicit the derivative asymptotic used in the radial angular-momentum obstruction. These are recorded in the review files and Git history. This is independent checking within an AI-assisted investigation, not independent human peer review or a substitute for the separate Lean checks described below. The proofs use standard Hilbert-space spectral theory, Lax--Milgram, Fredholm theory, Sobolev compactness and local elliptic estimates. The scalar scattering proof supplies its own outgoing uniqueness and material-Runge argument. Its elementary Bessel/Hankel facts can be checked against the Wronskian identities and large-order formulas in \cite{dlmf}. None of these background facts is a substitute for a Maxwell-to-operator bridge in the channel problem. The reproducible script \texttt{scripts/verify.py} checks the exact pair model, passive reciprocal perturbations, the angular monodromy, radial Bragg matching, the conical projection identities, and local evanescent flux. Its fifteen diagnostic groups use fixed seed $140014$ where random perturbations are needed. The additional modal groups check the exact sphere determinant against its rational exponential form, the first asymptotic coefficient and nine high-$Q$ dipole poles, and the constant-lifting cancellation through exponent one. The newest groups integrate the actual TE/TM radial systems for the growing-stack construction with step refinement, check the exact real-frequency dynamic-kernel bound, and independently integrate the conjugated angular generator through each layer of the identity loop. The last calculation includes both actual TE and TM interaction matrices. Another group computes source reaction coefficients from the actual regular/irregular transfer matrices of a common finite stack and checks their paired large-radius limit. They do not certify global channel synthesis or the continuous error bounds used in the proofs. All matrix tests use singular values or Hermitian eigenvalues, not tests of randomly selected coherent inputs. Random numbers choose one fully specified perturbation matrix; its full norm and all-subspace certificate are then evaluated. Floating-point residuals support consistency of the analytic proofs; they are not interval-certified error estimates. \begin{table}[ht] \centering \caption{Exact resonator-model diagnostic family, $J_j=10^{-j}$ and $\gamma_j=J_j$. This is not a Maxwell device simulation.} \label{ver:model-table} \begin{tabular}{@{}rrrrr@{}}\toprule $N$ & $\min J_j$ & $\sigma_{\min}(T)$ & maximum model energy & narrowest FWHM\\\midrule 2 & $10^{-2}$ & $1$ & $10^2$ & $2.82843\,10^{-2}$\\ 4 & $10^{-4}$ & $1$ & $10^4$ & $2.82843\,10^{-4}$\\ 8 & $10^{-8}$ & $1$ & $10^8$ & $2.82843\,10^{-8}$\\\bottomrule \end{tabular} \end{table} The ideal model's energy Gram matrix is $\operatorname{diag}(J_j^{-1})$ and its internal exchange-current Gram matrix is $I$; common detuning is zero. For weighted, purely Hamiltonian perturbations of norm $\rho=0.01$, Theorem~\ref{cx:n-model} bounds transmission-operator error by $0.0100713$. This is below $1-\sqrt{0.9}$, so it certifies transmission above $90\%$ in the perturbed model for every coherent vector. The reproduced smallest powers are approximately $0.999961$, $0.999959$, and $0.999949$. Model power-balance and full-unitarity residuals are below $10^{-14}$ in the reported run. The local evanescent example has an analytically exact section Gram matrix $I_8$. Direct integration of its explicit trigonometric fields with $33$, $65$ and $129$ transverse quadrature points gives operator errors below $10^{-14}$. This refines quadrature of a local analytic formula; it is not a domain, port, or material-mesh convergence study. The Bragg diagnostic uses $q=2$, $k_0=1$, $x_0\simeq2.74370727$ and material upper bound below $4.266$. For $M=0,2,4,8$ its core amplitude squared is approximately $8$, $115.292$, $1824.323$, $464888.834$, while reflection power remains one. These tests check normalization and exact matching, not an increasing transmission count. No numerical simulation of the analytic physical $N$-channel family in Theorem~\ref{uc:counterexample} was performed. Therefore numerical Maxwell transmission singular values, section-flux matrices, absolute bypass, weakest physical coupling, and relative model-reduction errors for $N=2,4,8$ are \emph{unavailable}, not zero. Their fields in the machine-readable record are null. No rounded-hotspot scattering mesh was computed; that theorem rests on the analytic existence and convergence proof, which is nonconstructive in its illumination cost and rounding scale. The supplied notes' claims about earlier finite-dimensional Lean checks were not rerun. A new formalization is described below, with its own source, pinned dependencies, successful compilation, and axiom audit. \subsection{Lean verification and its precise boundary} \label{ver:lean} The formalization uses Lean~4.19.0 with mathlib pinned to \begin{center}\small\texttt{c44e0c8ee63ca166450922a373c7409c5d26b00b}.\end{center} The source tree contains actual Lebesgue spaces, weak derivatives, compact test functions, distributional Maxwell equations, outgoing Green kernels and physical incoming and outgoing amplitudes. The existence theorems below construct these objects; their missing analytic assertions are not supplied as project axioms. \subsubsection*{Unrestricted power channels} The endpoint in \texttt{FullChannelExactPassiveWitness.lean}, \texttt{unrestricted\_exact\_passive}, takes only the fixed positive frequency, ordinary dielectric ceiling, square side, $0<\eta<\delta$, $0<\tau<1$, and finite dimension $N$. It constructs the actual finite shell-and-sphere material and an orthonormal physical incoming family. The counted space has dimension $N$. Its complete outgoing field, signed passage and whole-plane absolute bypass satisfy the stated inequalities for every coherent input, with the required flux integrability. The volume operator is proved invertible, and an actual bounded scattering map on transverse angular inputs is identified with the counted fields. Its restriction to the closed physical TE incoming Hilbert space satisfies $T^*T=I$, with complete outgoing radiation selected. This adjoint identity is attached to the same material and scattering map inside the exact physical witness. The corollary that no uniform finite count exists uses this constructed witness. No result for $\delta\leq\eta$ or for receiving-only output is asserted. \subsubsection*{Normalized modal hotspots} The endpoint \texttt{Packet06FinalAffineModal.lean}, \texttt{exists\_physical\_modal\_hotspots}, has no sharp-mode, excitation, regularity, collar or peak hypothesis. Its device certificate retains the literal nested material law and smooth interfaces, compact support, the actual complex volume kernel, reciprocal normalization one, the outgoing Maxwell field, and a genuine simple pole observed in the transported selected sector. All finite observation balls are disjoint, their projections lie in the requested patch, and actual continuous one-sided maxima lie on the prescribed affine plane. The same selected index retains open low collars and compact connected full half-amplitude components inside the balls. The material ceiling is the literal $\varepsilon_+>1$, and the observation balls have strict diameter below $\ell$. The construction uses an exact proper spatial frame and translation for every unit plane normal. Its spectral and physical transports are explicitly identified: the observed pole belongs to the same final material, support radius and actual normalized mode as the hotspots. The pole is simple in the transported selected sector; ambient rank-one simplicity is not asserted. The general electric-field identification theorem equates the potential representative used for the hotspots with the global outgoing field almost everywhere. \subsubsection*{Physical modal volumes} The endpoint \texttt{Packet06StrongAffineVolumePaper.lean}, \texttt{exists\_\allowbreak strong\_\allowbreak affine\_\allowbreak volume\_\allowbreak family}, constructs one common actual family for arbitrary finite $N$ in a prescribed affine-plane patch. It retains the literal dielectric ceiling, smooth three-value material, exact target real frequency, quality, normalization, outgoing behavior and the same selected-sector observed pole. Its closed observation balls are disjoint, project into the patch and have strict diameter below the requested length. The actual open collars are retained. The public wrapper instantiates the complete core construction at $K=\sqrt{\varepsilon_+}$; the separately checked scalar identity $K^2=\varepsilon_+$ gives the literal pointwise permittivity ceiling. For each weighted maximum, the same continuous phase representatives are certified against the actual field on a strictly larger closed ball before any maximum is chosen. These certificates survive dilation and the full spatial frame. The intrinsicness theorem proves that replacing them with any other such actual representatives leaves every smaller-ball trace value and the maximum unchanged, including at tangential interface contact. This strengthens the earlier small-ball-only predicates, which did not themselves justify that physical interpretation. The maximum uses the permittivity of the same material side at interfaces, and its reciprocal tends to zero. At these same selected weighted maxima, each tip has a fixed Cartesian component along a strictly increasing subsequence whose complex directional volume tends to zero. The exact normalized dilation factor is included in the proof. \subsubsection*{Driven finite-device hotspots} The endpoint \texttt{Packet07DrivenPlaneFinal.lean}, \texttt{plane\_final}, constructs a prescribed-frequency family in an arbitrary affine-plane patch, with one smooth transverse Herglotz illumination of physical incoming power one. The same actual electric and magnetic fields satisfy weak Maxwell and outgoing scattered-field limits. The material is binary, obeys the prescribed positive permittivity ceiling, has bounded support and vacuum exterior, and has smooth boundary and exactly $2N$ nonempty disjoint connected lobes. Disjoint closed observation balls have the prescribed strict diameter, and their projections and the support projection lie in the planar patch. For every finite list of thresholds, all sufficiently late members have simultaneous attained maxima and compact connected half-amplitude regions inside the balls. The maximizers themselves are not required to lie in the plane, in agreement with Theorem~\ref{hs:three-packing}. The exported Maxwell identity also yields actual $H(\operatorname{curl})$ fields on every bounded open region, with almost-everywhere equality to those same electric and magnetic fields; local square integrability of $E$, $H$ and $\varepsilon E$ is proved through the physical placement. \subsubsection*{Proof and build audit} These endpoints are assembled from the actual global representation, source-response identification, sphere spectral branch, nonzero conical excitation, normalized continuation, regional scalar transmission estimates and finite physical collar covers. The angular coefficient step uses a quantitative annular pairing; it does not assert analytic continuation of the intrinsic coefficient. The collar proof derives its mixed derivatives from the regional weak equation before using them in the localized forcing. It does not assume the unknown second derivatives on an outer cutoff annulus. Independent source review and separate direct-compiler builds accompany each accepted batch. The public entry point \texttt{OQ14Closure.lean} imports the channel, modal, volume and driven endpoints and prints their axiom dependencies. Exact ordered module lists, source and object hashes, compiler logs and reviews are retained under \texttt{checkpoints/}. Stage~116 contains 1586 incremental modules and 6001 standard-only reports beyond the earlier release. These are audit counts, not a measure of mathematical completeness. The aggregate project build passed 2171 modules and 10946 inline axiom reports. It excludes cached project objects, retaining earlier fresh objects only after source, log and object hash authentication. It records 14 historical linter warnings in four old modules; all new accepted batches have no diagnostics. A separate matching-stage audit checks every kernel-safe declaration owned by these project modules, including private and unprinted declarations: 23697 declarations passed. Unsafe executable compiler implementations are listed separately; the axiom collector does not filter dependencies of any checked declaration. This audit therefore covers more than the explicitly printed theorem reports. Lean and package objects remain trusted dependencies. Compiler executable hashes, the pinned mathlib revision and the audit tools are recorded, but this is not a from-source rebuild of Lean or mathlib. These checks establish the formal statements under that trusted environment; the manuscript-to- theorem correspondence is a separate source review. The only permitted logical axioms are \texttt{propext}, \texttt{Classical.choice} and \texttt{Quot.sound}. No admitted proof, project-specific mathematical axiom or numerical experiment replaces any step of the formal endpoint chains. The frozen earlier 5238-report release and its detailed batch history remain available as historical evidence; their earlier coverage limitations are superseded only where the new explicit endpoints discharge them. \subsection{Stricter questions outside the quantified targets} \label{ver:gaps} The unrestricted assertion has been disproved in Theorem~\ref{uc:counterexample} for every $\delta>\eta>0$ with full scattering output. The following resource estimates describe stronger restricted finiteness questions and the routes leading to that disproof. For a restricted channel proof, Theorem~\ref{ch:fulloutput} controls the complete physical output error after all feedback. What is missing is a uniform estimate of the effective transmission directions with no uncontrolled field-resource factor. A bound on $B_*$ on the counted subspaces would suffice for finiteness in the stated collar class, but the Bragg construction shows why external passivity cannot supply a general input-to-field bound. It does not rule out a sharper global argument using radiative accessibility, through-power and bypass together. Theorem~\ref{ca:localpackets} proves that local Maxwell equations and whole-plane coherent bypass control alone cannot supply the required restriction. The minimal abstract target is the uniform finite-rank approximation in Remark~\ref{ch:missing}; its area scaling is an additional requirement. Corollary~\ref{nr:channelcount} supplies a physical uniform resource bound when total dielectric volume is sufficiently small, without a support-diameter bound. That additional volume restriction is absent from the primary target. Theorem~\ref{cg:stack-count} separately gives an area bound in any prescribed finite angular band for the growing radial-stack construction: opposite-parity gains cancel in signed passage, even while local field energy diverges on large physical subspaces. Modes above the prescribed band remain outside that bound. Theorem~\ref{cn:radial-bypass} now excludes arbitrary high angular degeneracy in a centered homogeneous core, even with unlimited exterior resonant amplification and both polarizations. It leaves mixed angular orders and structured cores outside its scope. The earlier disproof criterion in Remark~\ref{cx:missing-bridge} specifies a quantitative alternative: a power-normalized Maxwell compression close to a unitary at one common frequency, plus the full physical flux inequality and absolute-bypass certificate on every coherent vector. Resonant algebra proves that weak coupling alone is not an obstruction, but it gives neither simultaneous bounded-index synthesis nor a relative error estimate for finite dielectric structures. Theorem~\ref{cn:local-synthesis} does provide actual bounded-index radial controls for small independent changes of finitely many complex core amplitudes. Its explicitly retained local range is not a global reachability or absolute-bypass theorem. Theorem~\ref{cc:global-control} now supplies approximate global reachability for every fixed finite set of TE orders, using amplified angular perturbations in distant positive-index layers. The complex core gains can be independently prescribed up to common phase. Theorem~\ref{cd:paired-control} also controls the paired TM transfers, and Theorem~\ref{cd:wall-isolation} amplifies one reflection sector while suppressing its opposite in a fixed $C_2$-even band. It does not estimate the response to a newly added inclusion or unselected angular orders. Separately, Theorem~\ref{cb:count} bounds counted dimension for any entire propagating Herglotz family using only its absolute bypass, without a gain bound. Corollary~\ref{cb:finitecore} transfers this obstruction to actual devices with a sufficiently large vacuum core at each fixed finite angular band. Its radius threshold may depend on the band. Thus neither result controls nearby material or arbitrary simultaneous growth of angular band and radius. Nearby material is essential to the subsequent counterexample. Theorem~\ref{cp:rank} demonstrates why it is essential: actual finite positive dielectrics can have arbitrary coherent passage rank and unit total throughput when bypass is omitted. Section~\ref{ct:section} constructs a local edge-circulation profile with equal passage and return, and exact finite-band jets with orthonormal physical inputs. Its Proposition~\ref{ct:completion} states the sufficient whole-plane $L^1$ Maxwell remainder. Theorem~\ref{cl:remainder} proves the self-consistent small-inclusion estimate for the explicit limiting kernel, and Theorem~\ref{rg:physical-limit} proves the all-order physical Green limit and whole-plane convergence for finite radial backgrounds. Theorem~\ref{uc:counterexample} combines them in the required order to disprove unrestricted finiteness for every fixed $\delta>\eta>0$. The selected output is the full scattering map; no claim is made here for stronger bypass thresholds or a prescribed receiving hemisphere. For hotspots, Theorem~\ref{hs:three-packing} supplies the finite-device driven construction directly, using conical points rather than an extrusion approximation. Its neighborhood projections lie in the prescribed planar patch; it does not require the rounded maxima to lie exactly in that plane. The separate modal construction now closes the modal bridge: Theorem~\ref{ms:modal-packing} produces one high-$Q$ resonance with a finite nonzero reciprocal residue norm, all singular coefficients nonzero, and smooth rounded peaks with actual maxima in the plane patch. The insertion contrast may depend on $N$, within the same fixed material bounds. The surrounding resonator's support is unrestricted. \subsection{Separate scientific verdicts} \label{ver:verdicts} \begin{longtable}{@{}p{.28\textwidth}p{.18\textwidth}p{.47\textwidth}@{}} \toprule Target & Status & Exact conclusion\\\midrule\endfirsthead \toprule Target & Status & Exact conclusion\\\midrule\endhead Driven, invariant 2D electric hotspots & PROVED & Theorem~\ref{hs:packing}: any finite number in any open patch, one physical unit-power-per-length illumination, fixed ordinary contrast and frequency, smooth rounded binary interfaces.\\[3pt] Driven finite 3D point hotspots & PROVED & Theorem~\ref{hs:three-packing}: one finite-power illumination, fixed small positive contrast, $2N$ smooth finite inclusions, disjoint 3D peak neighborhoods projecting into the prescribed patch.\\[3pt] Modal hotspot packing & PROVED & Theorem~\ref{ms:modal-packing}: one genuine normalized QNM, arbitrary finite peak count and quality factor, fixed real frequency and common ordinary bounds, actual maxima in the plane patch. Reciprocal residue normalization is explicit.\\[3pt] Scalar electric Helmholtz waves & OPEN here & The hotspot construction uses $u=H_z$ and in-plane vector $E$; it is not a singular scalar-electric-field example. Restricted scalar slab results do not settle OQ14.\\[3pt] Full vector, unrestricted fixed-section finiteness and area law & DISPROVED & Theorem~\ref{uc:counterexample}: arbitrary counted dimension at every fixed $\delta>\eta>0$, with the complete scattering map selected as output and common ordinary bounds. Stricter regimes remain separate.\\[3pt] Resource-controlled vector channels & PROVED / CONDITIONAL & Explicit finite-rank, count, average-resource and essential-norm results hold under the listed hypotheses. Uniform area scaling requires uniform additional resources.\\[3pt] Ordinary dielectrics & PROVED scoped results & Driven and modal hotspot constructions and Bragg/local-flux obstructions use common positive ordinary bounds. None uses ENZ.\\[3pt] Common positive lower bound below one & DISPROVED unrestricted count & Any such class admitting the scalar interval $[1,\eps_+]$ contains the counterexample of Theorem~\ref{uc:counterexample}. No ENZ limit is needed.\\[3pt] Shape-varying, ENZ, ideal guides & Excluded from primary target & The prior report supplies a qualified broader counterexample. Its varying lower bound and section shape are retained; this manuscript does not promote it or independently certify its entire construction.\\[3pt] Real symmetric positive anisotropy & PROVED local extension & Corollary~\ref{ch:anisotropy} extends the local resource theorem. Tensor classes admitting the scalar examples inherit both hotspot constructions and the unrestricted channel counterexample; stricter channel regimes are not settled.\\[3pt] Fixed total support & OPEN here & The growing-support Bragg sequence does not refute this more restricted problem.\\ Operator and Maxwell formalization & LEAN VERIFIED endpoints & Section~\ref{ver:lean} identifies the actual unrestricted-channel, arbitrary-plane modal and driven hotspot, and common arbitrary-plane modal-volume existence chains, with matching aggregate and all-safe-declaration audits.\\ \bottomrule \end{longtable} \subsection{Build and provenance} The canonical source is \texttt{manuscript/main.tex} and its included sections. The documented build is \texttt{python scripts/build.py} with Tectonic~0.17.0, followed or preceded by \texttt{python scripts/verify.py}. Tectonic runs the required bibliography and reference passes. The build script rejects unresolved final-pass citations, references and overfull boxes. Rendered pages are inspected separately; compilation never counts as mathematical verification. The build manifest records the source revision, working-tree state, engine, source hashes and PDF hash. The saved manuscript snapshot and its provenance are described in \texttt{paper/README.md}. The historical release record separately identifies its clean-checkout build and formal audit. These records do not establish independent human review.