\section{Unbalanced forms and vanishing on a singular stratum} \label{sec:zs-tools} We retain all eight coefficient directions, and we use the notation of Section~\ref{sec:balanced-alignment} (see the notation paragraph there; the letters $\iota$, $L$ and $G$ differ from their use in Sections~\ref{sec:real-fourfold-images} and~\ref{sec:u4-images}). Let $A,G$ be real Euclidean spaces of dimension eight and $V=E\oplus K$ an orthogonal sum of two eight-dimensional spaces. Fix an isometry $\iota:G\to K$, a nonzero $g_0\in G$, and an isomorphism $L:A\to E$. Let $M:A\to\operatorname{Hom}(G,V)$ be linear, and suppose \begin{equation}\label{eq:zs-setup} \operatorname{rank}M(a)=8\quad(a\ne0),\qquad M(a)g_0=-La,\qquad \iota^*M(a)\text{ is skew-adjoint}. \end{equation} Set $C(a)=M(a)M(a)^*$. Over the complex numbers, the adjoint denotes the transpose for the complex-bilinear extension, not the Hermitian adjoint. The earlier transpose construction supplies \eqref{eq:zs-setup}; the statements below are conditional linear-algebra tools and do not construct a normal form. We will use the following immediate consequence of the radial and skew identities. If all images of $M$ lie in one fixed space $T$ of dimension at most nine, then they equal $E$ for every nonzero real coefficient. Indeed, $L(A)=E\subset T$, so the projection of $T$ to $K$ has dimension at most one. The skew map $\iota^*M(a)$ has rank at most one and is zero. Its upper image then has dimension eight and equals $E$. \begin{proposition}[Unbalanced normal form]\label{prop:zs-unbalanced} Let $h,q$ be independent real quadratic forms on $A$, with $q$ positive definite. Suppose that $D_0,D_1$ are constant self-adjoint endomorphisms of $V$, $u\in V$ is a fixed nonzero vector, and $b:A\to V$ is quadratic, such that \[ C(a)=h(a)D_0+q(a)D_1+u b(a)^*+b(a)u^*. \] Then $\operatorname{Im}M(a)=E$ for every nonzero real $a$. \end{proposition} \begin{proof} Compress to $u^\perp$. The resulting matrix is $C_\perp(a)=h(a)B_0+q(a)B_1$, positive semidefinite of rank at most eight. On $q=1$, the Rayleigh quotient $h/q$ ranges over a nondegenerate closed interval $I$. Thus $B_1+tB_0\succeq0$ for $t\in I$. On the interior of $I$ its kernel is constant: expressing either of two interior parameters as a strict convex combination of the other and a third parameter proves this by the positive-semidefinite kernel-intersection identity. Let $F$ be the common image and $r=\dim F\le8$. Two distinct interior values show that both $B_0$ and $B_1$ annihilate $F^\perp$. Therefore $\operatorname{Im}C_\perp(a)\subset F$ for every $a$, including endpoint and exceptional values. Positivity of $C(a)$ puts its cross block in $\operatorname{Im}C_\perp(a)$: a vector in the kernel of a positive semidefinite principal block must be annihilated by the corresponding cross block. It follows that $\operatorname{Im}C(a)\subset\mathbb Ru+F$. Over the reals $\operatorname{Im}C(a)=\operatorname{Im}M(a)$. The fixed ambient space has dimension at most nine, so the preceding observation finishes. \end{proof} For the singular-stratum statement, choose coefficient coordinates $A=W\oplus Z$ with $\dim W=m\in\{4,5\}$ and \[ q(x,z)=|x|^2+|z|^2,\qquad h(x,z)=\lambda q(x,z)+\rho(z), \] where $\rho$ is a nondegenerate real quadratic form on $Z$. Its signature is unrestricted. Let \[ \mathcal A=\{(x,z)\in A_{\mathbb C}:q(x,z)=\rho(z)=0\}, \qquad \mathcal S=\{(x,0):|x|^2=0\}. \] We assume $\operatorname{rank}C\le2$ on $\mathcal A$. In the residue construction this follows from the generic rank bound on every component, since the three-minors are polynomials. In particular this assumption is not restricted to the smooth points of $\mathcal A$. \begin{proposition}[Vanishing on a multiplicity-four or -five stratum] \label{prop:zs-vanishing} In this setting, if $C$ vanishes on $\mathcal S$, then $\operatorname{Im}M(a)=E$ for every nonzero real $a$. \end{proposition} \begin{proof} We first align the restriction to $W$, then use complex arcs to constrain the remaining lower block. \emph{The restriction to $W$.} The quadric $|x|^2=0$ is irreducible over $\mathbb C$ for $m\ge4$. Every quadratic entry of $C(x,0)$ is therefore a scalar multiple of $|x|^2$. Write $C(x,0)=|x|^2D$, where $D$ is real positive semidefinite of rank eight, and put $F=\operatorname{Im}D$. For every nonzero real $x$, $\operatorname{Im}M(x,0)=F$. The radial identity puts $L(W)\subset F\cap E$, so the projection $R$ of $F$ onto $K$ has dimension at most $8-m$. The skew pencil $\iota^*M(x,0)$ has the same image $\iota^{-1}R$ for every nonzero real $x$. Its rank is even, and a skew map with image in $\iota^{-1}R$ is supported on that space. Rank two is impossible: skew maps on a fixed two-plane form a one-dimensional space, and a linear map from $W$ into that space cannot be nonzero at every nonzero $x$. The only other nonzero possibility is rank four when $m=4$. It would give an injective four-dimensional linear space of skew $4\times4$ matrices whose Pfaffian never vanishes away from zero. The restriction of the Pfaffian would be a definite quadratic form. But on the six-dimensional space of skew matrices the Pfaffian is \[ a_{12}a_{34}-a_{13}a_{24}+a_{14}a_{23}, \] which has signature $(3,3)$ and has no four-dimensional definite subspace. Thus $R=0$ and $F=E$. It follows that the lower pencil is $\iota^*M(x,z)=T(z)$, with $T$ real linear and skew-adjoint. The map $M(x,0):G\to E$ can be normalized on its \emph{output} by a fixed isomorphism $S:G\to E$ so that \begin{equation}\label{eq:zs-composition} M(x,0)=SO(x),\qquad O(x)^*O(x)=O(x)O(x)^*=|x|^2I,\qquad O(x_*)=I \end{equation} for a chosen real unit vector $x_*\in W$. For example, take $S=M(x_*,0)$, since $SS^*=D$. This normalization changes no coordinates on $G$, so $T$ remains skew-adjoint in the same domain as the endomorphisms $O(x)$. For $y\perp x_*$ set $J_y=O(y)$. Polarization gives \begin{equation}\label{eq:zs-clifford} J_y^*=-J_y,\qquad J_yJ_{y'}+J_{y'}J_y=-2\langle y,y'\rangle I. \end{equation} \emph{Complex arcs and exterior squares.} Fix any nonzero complex $z_0$ with $\rho(z_0)=0$, and any nonzero complex $x_0$ with $|x_0|^2=0$. The differential of $|x|^2$ at $x_0$ is nonzero, so there is a local complex analytic arc $x(t)=x_0+O(t^2)$ satisfying $|x(t)|^2=-t^2|z_0|^2$. Then $a(t)=(x(t),tz_0)$ lies in $\mathcal A$. Since $C(a(0))=0$, the coefficient of $t$ in $C(a(t))$ is \[ C_1=M(x_0,0)M(0,z_0)^*+M(0,z_0)M(x_0,0)^*. \] Every three-minor of $C(a(t))$ is zero, so its coefficient of $t^3$ gives $\operatorname{rank}C_1\le2$. Relative to $E\oplus K$ the lower diagonal block of $C_1$ is zero, while the cross block is $M(x_0,0)T(z_0)^*$, after identifying $K$ with $G$. The rank of a symmetric block matrix $\left[\begin{smallmatrix}A&B\\ B^*&0\end{smallmatrix}\right]$ is at least $2\operatorname{rank}B$. Hence \begin{equation}\label{eq:zs-null-collapse} \operatorname{rank}\bigl(O(x_0)T(z_0)^*\bigr)\le1. \end{equation} Suppose $T(z_0)\ne0$, and put $U=\operatorname{Im}T(z_0)^*\subset G_{\mathbb C}$. Its dimension $r$ is positive and even. A fixed complex right inverse onto $U$ transfers \eqref{eq:zs-null-collapse} to $O(x_0)|_U$. The entries of $\bigwedge^2O(x)|_{\bigwedge^2U}$ are quadratic and vanish on $|x|^2=0$, so they are constant multiples of $|x|^2$. Evaluation at $x_*$ identifies those constants: \[ \left.\bigwedge\nolimits^2O(x)\right|_{\bigwedge^2U} =|x|^2\,\mathrm{inc}_{\bigwedge^2U}. \] Differentiating at $x_*$ in a real direction $y\perp x_*$ yields \begin{equation}\label{eq:zs-wedge-derivative} J_yu\wedge v+u\wedge J_yv=0\qquad(u,v\in U). \end{equation} Projection to the terms with one factor outside $U$ first shows that $J_yU\subset U$: for independent $u,v$, their distinct $U$ factors force both outside components to vanish. If $r\ge3$, the induced action of an endomorphism of $U$ on $\bigwedge^2U$ is faithful. In a basis, a third basis vector detects every off-diagonal entry; the diagonal entries satisfy $d_i+d_j=0$ for every pair and therefore all vanish. Equation~\eqref{eq:zs-wedge-derivative} would give $J_y|_U=0$, contrary to $J_y^2=-|y|^2I$ for any nonzero real $y$. Thus $r=2$, and \eqref{eq:zs-wedge-derivative} says \begin{equation}\label{eq:zs-traceless} \operatorname{tr}(J_y|_U)=0. \end{equation} This argument is over the full complex endomorphism algebra; it does not require the bilinear form on $U$ to be nondegenerate. \emph{Multiplicity five.} If $m=5$, choose four orthonormal real vectors in $x_*^\perp$ and denote their operators by $J_1,\ldots,J_4$. On the invariant complex two-space $U$, equations \eqref{eq:zs-clifford} and \eqref{eq:zs-traceless} give \[ \operatorname{tr}(J_iJ_j|_U)=0\ (i\ne j),\qquad \operatorname{tr}(J_i^2|_U)=-2. \] The four restrictions are therefore linearly independent, although they lie in the three-dimensional space of traceless $2\times2$ complex matrices. This contradiction proves $T(z_0)=0$ for every complex $\rho$-null $z_0$. The null cone of the nondegenerate rank-three form spans $Z_{\mathbb C}$, so linearity gives $T=0$. \emph{Multiplicity four.} If $m=4$, there are three such operators $J_1,J_2,J_3$. For a complex null $z_0$ with nonzero $T(z_0)$, its rank-two image $U$ is invariant and each $J_i|_U$ has trace zero. A skew map with image $U$ corresponds to a bivector in $\bigwedge^2U$. The induced action of $J_i$ on that line is its trace and is zero. In matrix terms this gives \[ J_iT(z_0)+T(z_0)J_i^*=J_iT(z_0)-T(z_0)J_i=0. \] The same equation holds if $T(z_0)=0$. Each commutator is linear in $z$, and the nondegenerate rank-four null cone spans $Z_{\mathbb C}$. Consequently $T(z)$ commutes with all three $J_i$ for every $z$. The actual radial identity in \eqref{eq:zs-setup} gives $T(z)g_0=0$. Commutation with $J_1,J_2$ makes $T(z)$ zero on the real four-space \[ H=\operatorname{span}_{\mathbb R} \{g_0,J_1g_0,J_2g_0,J_1J_2g_0\}. \] These vectors divided by $|g_0|$ are orthonormal by \eqref{eq:zs-clifford}. The space $H$ and its orthogonal complement are invariant under $J_1,J_2$. Skewness makes $T$ preserve $H^\perp$. On this four-dimensional complement, $J_1,J_2$ give the left action of the quaternions on a quaternionic line. A commuting real endomorphism is right multiplication by a quaternion; skew-adjointness restricts it to an imaginary quaternion. Explicitly, choose a unit $v\in H^\perp$. A commuting map is determined by its value at $v$ in the basis $v,J_1v,J_2v,J_1J_2v$, and skewness removes its $v$ coefficient. Hence there are three real linear forms $t_1,t_2,t_3$ on $Z$ such that \begin{equation}\label{eq:zs-quaternion-square} \left.T(z)^2\right|_{H^\perp} =-\tau(z)I,\qquad \tau=t_1^2+t_2^2+t_3^2. \end{equation} We have already proved $\operatorname{rank}T(z)\le2$ whenever $\rho(z)=0$ over $\mathbb C$. If $\tau(z)\ne0$ at such a point, \eqref{eq:zs-quaternion-square} would make the restriction to $H^\perp$ invertible, of rank four. Therefore $\tau$ vanishes on the irreducible rank-four quadric $\rho=0$. Both polynomials have degree two, so $\tau=c\rho$ for a real constant $c$. The rank of $\tau$ is at most three, whereas $\rho$ has rank four, forcing $c=0$. Over the reals a sum of three squares can vanish identically only if each $t_i$ is zero. Thus $T=0$ also when $m=4$. In both cases the lower projection of $M$ is zero. Its real rank eight identifies its image with $E$, completing the proof. \end{proof} \begin{remark}\label{rem:zs-boundaries} The proof uses complex transverse null directions even when $\rho$ is definite over the reals. It never assumes that their rank-two image spaces are real or nondegenerate. The multiplicity-four finish uses the actual radial kernel vector, in addition to the Clifford relations. No integrality or divisor-class argument is needed once the stated rank bound on every component and vanishing on $\mathcal S$ are known. If $C$ has generic rank two on that stratum instead, extracting global real normal forms is a separate task. Finally, alignment at one anchor is the conclusion here; the established real-open propagation and closed-kernel criterion supply the later Clifford conclusion when these hypotheses hold along a branch of anchors. \end{remark}