\section{Preliminary reductions for the conjugate five-dimensional images} \label{sec:m6-preliminary} Let $B(a)$ be the actual real linear transpose pencil at a regular rank-eight anchor, and write $G=B B^T$. For every nonzero real coefficient $a$, it has rank eight. Its radial column is $B(a)e=-\iota(a)$, where $\iota$ is an isomorphism onto the physical Jacobi image $E$, and $K^TB(a)$ is skew for the orthogonal lower frame $K$. A real change of coefficient coordinates is allowed: no isometry property of the resulting radial map $\iota$ is used. Proposition~\ref{prop:db-positive-model} supplies real coordinates $a=(x,z)\in W\oplus Z$, with $\dim W=6$, $\dim Z=2$, and \begin{equation}\label{eq:m6-positive-model} q(x,z)=|x|^2+|z|^2,\qquad h(x,z)=\lambda q(x,z)+\rho(z). \end{equation} Here $q$ is positive definite, $h$ is nondegenerate indefinite, and $\rho$ is a definite nondegenerate binary form. Write $\rho$ as a nonzero real multiple of $\ell\overline\ell$, with $\ell$ the complex linear form of Proposition~\ref{prop:db-single}(2). Over $\mathbb C$ the two null lines of $\rho$ give distinct conjugate components \[ A_+=V(h,\ell),\qquad A_-=V(h,\overline\ell),\qquad A_+\cup A_-=V(h,q)=V(q,\rho). \] This union is a reduced complete intersection: $(q,\ell)$ and $(q,\overline\ell)$ are prime, and in the factorial ring $\mathbb C[a]/(q)$ \cite[Theorem~8.2, p.~26]{Samuel1964} the nonassociate primes $\ell,\overline\ell$ satisfy $(\ell)\cap(\overline\ell)=(\ell\overline\ell)$. Hence the ideal of $A_+\cup A_-$ is $(q,\rho)=(h,q)$. Each component spans its nonreal hyperplane and is a quadric of rank six or seven. The Gram has rank at most two everywhere on the union, generically two on each component. These geometric and rank statements are consequences of that proposition, Remark~\ref{rem:db-scope} and Lemmas~\ref{lem:si-incidence}, \ref{lem:si-saturated-fibres} and~\ref{lem:cp-complex-ranks} in type $(\alpha,s)=(5,5)$; by the last lemma the rank of the Gram is even at every point. Set \[ \Sigma=A_+\cap A_-=\{z=0,\ |x|^2=0\}. \] \subsection{Vanishing on the intersection} \begin{proposition}\label{prop:m6-zero-intersection} If $G$ vanishes identically on $\Sigma$, then $\operatorname{im}B(a)=E$ for every nonzero real $a$. \end{proposition} \begin{proof} The complex quadric $|x|^2=0$ in six variables is irreducible. Thus \[ G(x,0)=|x|^2D \] for a constant real positive semidefinite matrix $D$ of rank eight. All nonzero real $x\in W$ give the same image $F=\operatorname{im}D$. The radial column puts a six-space in $F\cap E$, so the lower projection of $F$ has dimension at most two. For every such $x$ the lower skew map has precisely that image. Rank two would give a linear map from six parameters to the one-dimensional skew space on a fixed two-plane with no nonzero kernel, which is impossible (a skew map has even rank, so dimension one is impossible). Hence $F=E$ and $K^TB(x,z)=T(z)$ is a linear skew pencil in $z$ alone. Choose an isomorphism $S:\R^8\to E$ with $SS^T=D$. The real square pencil $O(x)=S^{-1}B(x,0)$ satisfies \begin{equation}\label{eq:m6-norm-pencil} O(x)O(x)^T=O(x)^TO(x)=|x|^2I_8. \end{equation} Fix any nonzero complex $z_0$ with $\rho(z_0)=0$ and any nonzero complex $x_0$ with $|x_0|^2=0$. The nonzero gradient in the $x$ variables gives a formal or analytic arc $x(t)=x_0+O(t^2)$ satisfying \[ |x(t)|^2+t^2|z_0|^2=0. \] Throughout, $|\cdot|^2$ on complex vectors denotes the complex-bilinear extension of the real quadratic form. The arc $(x(t),tz_0)$ lies in $V(q,\rho)$, and the Gram vanishes at $t=0$. Its coefficient of $t$ is \[ G_1=B(x_0,0)B(0,z_0)^T+B(0,z_0)B(x_0,0)^T. \] The three-minors vanish along the arc, so their leading coefficients give $\rank G_1\leq2$. Its lower diagonal block is zero and its upper--lower block is $B(x_0,0)T(z_0)^T$. Since a symmetric block matrix $\left[\begin{smallmatrix}A&C\\ C^T&0\end{smallmatrix}\right]$ has rank at least $2\rank C$, we obtain \begin{equation}\label{eq:m6-null-collapse} \rank\bigl(O(x_0)T(z_0)^T\bigr)\leq1. \end{equation} This argument is valid on either complex transverse null line. Suppose $T(z_0)\ne0$, and put $U=\operatorname{im}T(z_0)^T$. Its complex dimension $r$ is positive and even. A constant right inverse onto $U$ transfers \eqref{eq:m6-null-collapse} to $O(x_0)|_U$. Its quadratic two-minors vanish on the irreducible quadric $|x|^2=0$, so they are constant multiples of $|x|^2$. Normalize at a real unit $x_*$ by replacing $O(x)$ with $O(x_*)^TO(x)$. Evaluation at $x_*$ then yields the identity \begin{equation}\label{eq:m6-wedge-identity} \bigwedge\nolimits^2O(x)\big|_{\wedge^2U} =|x|^2\,\operatorname{incl}_{\wedge^2U}. \end{equation} For real $y\perp x_*$, let $J_y=O(y)$. Polarization of \eqref{eq:m6-norm-pencil} gives $J_y^T=-J_y$ and $J_y^2=-|y|^2I$. Differentiating \eqref{eq:m6-wedge-identity} gives \[ J_yu\wedge v+u\wedge J_yv=0\qquad(u,v\in U). \] The components outside $U$ show that $J_yU\subset U$. If $r\geq3$, the infinitesimal exterior-square action of $\operatorname{End}(U)$ is faithful: a third basis vector detects each off-diagonal entry, while diagonal entries satisfy $d_i+d_j=0$ for all pairs, forcing all to vanish. Thus $J_y|_U=0$, contrary to its square for nonzero $y$. If $r=2$, the same identity says $\tr(J_y|_U)=0$. Choose five real orthonormal vectors $y_i\in x_*^\perp\cap W$ and put $J_i=J_{y_i}$. Polarization gives \[ J_iJ_j+J_jJ_i=-2\delta_{ij}I, \qquad \tr\bigl((J_i|_U)(J_j|_U)\bigr)=-2\delta_{ij}. \] The five restrictions are therefore complex linearly independent, but all lie in the three-dimensional space of traceless two-by-two complex matrices. This contradiction proves $T(z_0)=0$. No nondegeneracy or reality of the subspace $U$ was assumed. The two distinct complex null lines of the nondegenerate binary form $\rho$ span $Z_{\mathbb C}$. Since $T$ is linear and vanishes on both, it is zero. Real rank eight now gives the claimed image. \end{proof} \subsection{A constant component column} \begin{proposition}\label{prop:m6-constant-column} Suppose that on $A_+$ one has \[ G=u b^T+b u^T, \] where $u$ is a fixed nonzero complex physical vector and $b$ is a homogeneous quadratic vector in the component coordinate ring. Then $\operatorname{im}B(a)=E$ for every nonzero real $a$. \end{proposition} \begin{proof} Let $F=\Span_{\R}\{\operatorname{Re}u,\operatorname{Im}u\}$ and let $\Pi$ be real orthogonal projection to $F^\perp$. Then $\dim F\leq2$. The compressed Gram $\Pi G\Pi$ vanishes on $A_+$, and reality makes it vanish on $A_-$ as well. The reduced union has degree-two ideal $\Span\{h,q\}$. Therefore \[ C(a):=\Pi G(a)\Pi=h(a)D_0+q(a)D_1 \] for constant real symmetric matrices. It is positive semidefinite of rank at most eight at every real coefficient; constant rank is not needed. On $q=1$, the Rayleigh range of $h$ is a nontrivial interval $[\alpha,\beta]$. Every $tD_0+D_1$ on that interval is positive semidefinite. At interior parameters it is a strict positive combination of the endpoint matrices, so all interior kernels are their common intersection. The corresponding fixed image $V_0$ has dimension at most eight. On the nonempty real open set of interior Rayleigh values, $C=(\Pi B)(\Pi B)^T$ has image $V_0$. Projection of the linear pencil $\Pi B$ to $V_0^\perp$ vanishes there, and therefore identically. All images of $B$ lie in the fixed real space $F+V_0$ of dimension at most ten. Lemma~\ref{lem:m03-ten-space} applies to this reduced union. To check its component hypothesis, a nonzero real linear form vanishing on $A_+$ would vanish on its span $\ker\ell$, hence would be a complex multiple of $\ell$. Reality would make it a multiple of $\overline\ell$ as well, impossible because the two forms are independent. The same holds for $A_-$. This proves the conclusion. In particular, no constancy of the compressed rank at the Rayleigh endpoints has been inserted into the argument. \end{proof} \subsection{The remaining component columns are linear} We recall the actual-pencil pairing input for the following extraction. At a generic point of $A_+$, the kernel of $B$ is a totally isotropic three-space in its standard eight-dimensional domain. The induced rank-two symmetric pairing is the binary quotient of that form by three hyperbolic planes; over the complex component field it is hyperbolic. This is the isotropic-kernel identity of Lemma~\ref{lem:cp-complex-ranks}: over the component field, $K_0^\perp/K_0$ has determinant class $-1$, and $-1$ is a square there. It is not a consequence of a rank-two matrix alone. \begin{proposition}\label{prop:m6-linear-extraction} If the alignment conclusion of Proposition~\ref{prop:m6-constant-column} does not already hold, then on $A_+$ \[ G=p r^T+r p^T \] with complex homogeneous linear columns $p,r$. On $A_-$ the factorization is their conjugate. \end{proposition} \begin{proof} The coordinate ring $S_+=\mathbb C[a_1,\ldots,a_8]/(h,\ell)$ is a UFD: the restricted quadric has rank six or seven, so Klein--Nagata \cite[Theorem~8.2, p.~26]{Samuel1964} applies, with a polynomial extension if there is a radical variable. The two lines are stable under coefficient scaling, so their saturated modules are graded and have homogeneous generators, as in the proof of Lemma~\ref{lem:pf-principal-extraction}. Thus the two homogeneous isotropic lines of the generic rank-two pairing have primitive homogeneous polynomial generators $p,r$ in this ring. Equivalently, their saturated rank-one reflexive modules are principal over the UFD. Thus over its fraction field \[ G=t(p r^T+r p^T) \] for a homogeneous scalar $t$. At a height-one prime, each primitive vector has a nonzero reduction. Over a field of characteristic zero, the symmetric cross product of two nonzero vectors cannot be the zero matrix, even if the vectors are proportional: choose a nonzero coordinate in the first vector and use its diagonal and row entries. Hence some entry of the cross product is a unit at that prime. Regularity of $G$ forces the valuation of $t$ to be nonnegative there. Factoriality therefore gives $t\in S_+$. The nonnegative degrees satisfy \[ \deg t+\deg p+\deg r=2. \] If either column has degree zero, absorb $t$ and the other column into the quadratic vector of Proposition~\ref{prop:m6-constant-column}. Otherwise both column degrees are one and $t$ is constant. Absorbing this nonzero complex constant into one column gives the stated form. \end{proof} \begin{remark}\label{rem:m6-input-boundaries} The conclusions in this section are pointwise image statements or componentwise factorizations. Generic-anchor closed-kernel propagation remains a separate input. Likewise, the two complex linear columns must still be glued to a real global identity, and alignment of a real balanced form is a separate criterion. No extension theorem on the reducible, nonnormal union is used here. A component column may acquire a common factor when restricted to $\Sigma$; the later real-gluing argument must handle that possibility. \end{remark}