\section{Alignment of a balanced quadratic normal form} \label{sec:balanced-alignment} We isolate the linear algebra needed after a real balanced normal form has been constructed. All coefficient directions are retained in this argument. \emph{Notation.} In this section and in Section~\ref{sec:zs-tools}, the pencil is written $M(a)$, its Gram matrix is $C(a)=M(a)M(a)^*$, $\iota$ is an isometry onto the kernel space $K$, and $L$ is the radial isomorphism onto $E$. In Sections~\ref{sec:real-fourfold-images} and~\ref{sec:u4-images} the same objects are written $B(a)$, $G(a)=B(a)B(a)^T$, $K$ and $\iota$, and $L(a)$ there denotes the lower skew block $K^TB(a)$. Let $A$ and $G$ be eight-dimensional Euclidean spaces, and let $V=E\oplus K$ be an orthogonal decomposition with $\dim E=\dim K=8$. Fix an isometry $\iota:G\longrightarrow K$, a nonzero vector $g_0\in G$, and an isomorphism $L:A\longrightarrow E$. Put $x=\iota g_0$. Consider a linear map \[ M:A\longrightarrow\operatorname{Hom}(G,V). \] We assume that, for all $a\in A$, \begin{equation}\label{eq:balanced-radial-skew} M(a)g_0=-La,\qquad \iota^*M(a)\text{ is skew-adjoint}, \end{equation} and that \begin{equation}\label{eq:balanced-nonsingular} N_g:A\longrightarrow V,\qquad N_g(a)=M(a)g, \quad\text{is injective for every }g\ne0. \end{equation} In particular, $M(a)$ has rank eight for every $a\ne0$: a nonzero vector in its kernel would contradict \eqref{eq:balanced-nonsingular}. Thus $C(a)=M(a)M(a)^*$ is positive semidefinite of rank eight. The two identities in \eqref{eq:balanced-radial-skew} also give \begin{equation}\label{eq:balanced-tangency} \operatorname{Im}M(a)\subset x^\perp, \qquad \langle\iota g,M(a)g\rangle=0. \end{equation} Indeed, $\iota^*M(a)g_0=0$, so skew-adjointness makes $g_0$ orthogonal to the image of $\iota^*M(a)$. \begin{lemma}\label{lem:balanced-small-ambient} If all images $\operatorname{Im}M(a)$ lie in one fixed subspace $T\subset V$ of dimension at most nine, then $\operatorname{Im}M(a)=E$ for every $a\ne0$. \end{lemma} \begin{proof} The radial identity gives $E=L(A)\subset T$. Consequently the projection of $T$ onto $K$ has dimension at most one. The skew-adjoint map $\iota^*M(a)$ therefore has rank at most one, and hence is zero. This puts every image in $E$; its rank is eight. \end{proof} \begin{proposition}\label{prop:balanced-alignment} Suppose that $q$ is a positive definite quadratic form on $A$ and $h$ is a real quadratic form. Let $\lambda_1\le\cdots\le\lambda_8$ be the eigenvalues of $h$ relative to $q$. Assume there are constant self-adjoint endomorphisms $D_0,D_1$ of $V$ and real linear maps $u,v:A\longrightarrow V$ such that \begin{equation}\label{eq:balanced-normal-form} C(a)=h(a)D_0+q(a)D_1+u(a)u(a)^*-v(a)v(a)^*. \end{equation} If $\lambda_2<\lambda_7$, then $\operatorname{Im}M(a)=E$ for every nonzero real $a$. \end{proposition} \begin{proof} Write $D(t)=D_1+tD_0$. Fix $b\in V$ and impose the single linear condition $\langle b,u(a)\rangle=0$. Its solution space $A_b\subset A$ has codimension at most one. On $A_b\setminus\{0\}$, positivity of \eqref{eq:balanced-normal-form} gives \[ q(a)\langle D(h(a)/q(a))b,b\rangle \ge \langle b,v(a)\rangle^2\ge0. \] The minimum of $h/q$ on $A_b$ is at most $\lambda_2$, and its maximum is at least $\lambda_7$, by the min--max principle. Its Rayleigh range is an interval, so the displayed inequality proves \begin{equation}\label{eq:balanced-positive-interval} D(t)\succeq0\quad\text{for every }t\in[\lambda_2,\lambda_7]. \end{equation} Here $b$ was arbitrary; the vector $a$ used to test a given $t$ may depend on $b$. On the interior of this interval, $D(t)$ has one common kernel and therefore one common image $F$. To see this, express one interior parameter as a strict convex combination of another and a third parameter in the interval. The kernel of a positive convex combination of two positive semidefinite matrices is the intersection of their kernels. Reversing the two interior parameters proves equality. Let $r=\dim F$. The set \[ \Omega=\{a\ne0:\lambda_2