\section{Closed kernels and a fixed ten-dimensional image} \label{sec:m03-closed-kernels} Throughout this section, $V=\mathbb R^{16}$ is Euclidean and $\Jac_x$ is the Jacobi family of a shifted Osserman algebraic curvature tensor of rank eight. Write $q(x)=\langle x,x\rangle$, and extend all polynomial identities complex linearly. The distinct nonzero reduced eigenvalues are the constants $\lambda_1,\ldots,\lambda_t$ times $q(x)$. For a real nonzero anchor $x$, put $E_x=\operatorname{Im}\Jac_x$ and $K_x=\ker\Jac_x$. We use the coherent-partner construction established earlier. At a regular real unit anchor $e$, choose orthonormal image and kernel frames, with image frame $M_e$ and kernel frame $K_e$. In these frames, the rational factor at $z=te+y$, $y\perp e$, has the form \begin{equation}\label{eq:m03-partner} M_z=\begin{bmatrix}tI+F(y)\\P(y)\end{bmatrix},\qquad \Jac_z=M_z\Lambda M_z^T, \end{equation} where $\Lambda$ is an invertible constant diagonal matrix, $F$ is rational, and $P$ is a generically invertible linear $8\times8$ matrix. This construction is an input to the lemmas below. \begin{lemma}[Closed-kernel criterion]\label{lem:m03-closed-kernel} Suppose that, on a nonempty Euclidean-open set of real anchors $x$, \begin{equation}\label{eq:m03-ck} \Jac_y z=0\qquad\text{for every }y,z\in K_x. \end{equation} Then $\Jac_x^2=0$ for every complex $x$ with $q(x)=0$. Moreover, at every regular real anchor $e$, the full coherent-partner Gram matrix is radial: $M_{te+y}^TM_{te+y}=(t^2+q(y))I$ as a rational identity. \end{lemma} \begin{proof} We first check that the complex null quadric contains a dense set of rank-eight Jacobi operators. Over $\mathbb C(y)$, adjoin $\alpha$ with $\alpha^2=-q(y)$. The polynomial $q(y)$ has rank fifteen, so it is not a square up to sign in this rational function field. The lower block $P(y)$ in \eqref{eq:m03-partner} remains invertible after this extension. Consequently $M_{\alpha e+y}$ is injective and its transpose is surjective. Since $\Lambda$ is invertible, $\Jac_{\alpha e+y}$ has rank eight. This assertion requires no nondegeneracy of its Gram matrix. Every nine-minor of $\Jac_x$ vanishes identically. On its rank-eight locus, choose a rational kernel basis using a nonzero eight-minor. Substitution of arbitrary linear combinations of this basis for $y,z$ in \eqref{eq:m03-ck} gives finitely many rational identities in $x$. They vanish on a nonempty real open subset, hence identically. Such minor charts cover the generic null locus just found. Thus \eqref{eq:m03-ck} holds at its generic complex points. In particular, we have not evaluated a spectral projector with denominator $q(x)$ on the null quadric. Fix such a null point $x$, and set $K=\ker\Jac_x$. Then $\dim K=8$, and the Osserman characteristic polynomial makes $\Jac_x$ nilpotent. If $q|_K$ is not identically zero, choose $y\in K$ with $q(y)\ne0$. The continued closed-kernel identity gives $\Jac_yK=0$. The polynomial identity \[ \Jac_y\prod_{j=1}^t(\Jac_y-\lambda_jq(y)I)=0 \] holds over $\mathbb C$, by continuation from the real family. Its roots are distinct when $q(y)\ne0$, so $\Jac_y$ is diagonalizable. Its characteristic polynomial gives rank eight, whence $K=\ker\Jac_y$. The kernel of a self-adjoint diagonalizable operator on a nondegenerate bilinear space is nondegenerate. Hence $K$ is nondegenerate. But self-adjointness of $\Jac_x$ then gives $\operatorname{Im}\Jac_x=K^\perp$, and $\Jac_x$ restricts to an invertible operator on $K^\perp$, contradicting nilpotency. Therefore $q|_K=0$. Polarization makes $K$ totally isotropic, and its dimension gives $K=K^\perp$. It follows that $\operatorname{Im}\Jac_x=\ker\Jac_x$, and hence $\Jac_x^2=0$. The null quadric is irreducible, so this polynomial identity extends to all its points, including those of smaller Jacobi rank. Finally put $G_z=M_z^TM_z$. At $z=\alpha e+y$, \[ 0=\Jac_z^2=M_z\Lambda G_z\Lambda M_z^T \] implies $G_z=0$, by left and right inverses. Expanding the Gram matrix in the radial variable gives \[ G_{te+y}=t^2I+tS(y)+\Phi(y). \] Thus $\alpha S(y)+\Phi(y)-q(y)I=0$. Independence of $1,\alpha$ over $\mathbb C(y)$ yields $S=0$ and $\Phi=q(y)I$, as required. \end{proof} \begin{remark}\label{rem:m03-scope} The last lemma supplies the full radial Gram, not just its restriction to $e^\perp$. Applying the previously established radial-Gram classification is a separate step. In the remaining nonscalar patterns it gives a Clifford structure; the scalar two-root Cayley alternative is handled separately. \end{remark} For completeness, the connection with the transpose pencil is direct. At a real anchor, define $N_g$ by $N_g^Ty=P(y)^Tg$, extending $P$ to vanish in the anchor direction. Sectional symmetry gives $\Jac_{K_eg}=N_g\Lambda N_g^T$. If the full transpose pencil $B(a)g=N_ga$ has image contained in $E_e$ for every coefficient $a$, then $\operatorname{Im}N_g\subset E_e$. It follows that $\Jac_{K_eg}$ kills $K_e$, proving the closed-kernel condition at this anchor. Applying this on a real open set permits Lemma~\ref{lem:m03-closed-kernel}. One may also propagate this condition away from the null cone using the rational projector \[ \Pi_0(x)=\prod_{j=1}^t \frac{\Jac_x-\lambda_jq(x)I}{-\lambda_jq(x)} \] and the identity $\Jac_{\Pi_0(x)y}\Pi_0(x)z=0$. Clearing denominators makes propagation frame independent; the minor charts in the proof above supply the subsequent passage to null points. \begin{lemma}[Fixed ten-dimensional image]\label{lem:m03-ten-space} Fix a regular real anchor $e$. Let $B(a)$ be its actual linear transpose pencil, with eight real coefficient variables, and set $C(a)=B(a)B(a)^T$. Thus $C(a)$ is positive semidefinite of rank eight for every real $a\ne0$. Suppose that all images of $B(a)$ lie in a fixed real subspace $T$ of dimension at most ten. Suppose also that the coefficient residue locus is a reduced complete intersection $A=V(h,q_0)\subset\mathbb P^7_{\mathbb C}$, where $h,q_0$ are independent real quadrics and $q_0$ is positive definite. Assume that $C$ has generic rank two on each component of $A$, and that no nonzero real linear form vanishes identically on any component. Then $\operatorname{Im}B(a)=E_e$ for every real $a\ne0$. \end{lemma} \begin{proof} The radial column $B(a)g_e=-M_ea$ puts $E_e\subset T$. The lower projection $\Gamma(a)=K_e^TB(a)$ is skew-symmetric and has image in a fixed real space $R$ of dimension at most two. If $\dim R$ is at most one, skewness gives $\Gamma=0$, and the assertion follows from rank eight. Otherwise, every such skew map is a multiple of a fixed rank-two skew map. Write $\Gamma(a)=\ell(a)J$, with $\ell$ real linear. We may assume $\ell\ne0$. On $E_e\oplus R$ the Gram has block form \[ C(a)=\begin{bmatrix} A_0(a)&\ell(a)L(a)\\ \ell(a)L(a)^T&\ell(a)^2D \end{bmatrix}, \] where $D$ is constant positive definite of size two and $L$ is linear. These degrees follow from the linear factor $B$ and its lower block $\ell J$. The component hypothesis ensures that $\ell$ is nonzero generically on each component of $A$. There the lower block has rank two, so the rank-two condition on $C$ makes its Schur complement zero. Consequently the quadratic matrix \[ H=A_0-LD^{-1}L^T \] vanishes on the reduced complete intersection. Its degree-two ideal is spanned by $h,q_0$, and therefore $H=hH_0+q_0H_1$ with constant real symmetric matrices $H_0,H_1$. At real points with $\ell\ne0$, the Schur complement is positive semidefinite of rank six. Continuity gives positivity and rank at most six everywhere. At real nonzero points of $\ker\ell$, the image of $B$ lies in $E_e$, so $A_0$ is positive definite of rank eight. Since $LD^{-1}L^T$ has rank at most two, $H$ has rank at least six there as well. Thus $H$ has constant rank six for all real nonzero coefficients. Normalize $q_0=1$. The range of $h$ is a nontrivial closed interval, and the pencil $tH_0+H_1$ is positive semidefinite of rank six on that interval. The kernel of a strict positive combination of two positive semidefinite matrices is the intersection of their kernels. Constancy of rank therefore makes these kernels equal, including at the endpoints. Hence all $H(a)$ have one fixed six-dimensional image $F\subset E_e$. On the seven-dimensional real space $\ker\ell$, compress $A_0=H+LD^{-1}L^T$ to $F^\perp\cap E_e$, a two-dimensional space. The result is $L_2(a)D^{-1}L_2(a)^T$, positive definite for every nonzero $a\in\ker\ell$. Thus the real $2\times2$ matrix $L_2(a)$ would be invertible at every such point. But $L_2$ is a linear map from a seven-dimensional space into a four-dimensional matrix space, so it has a nonzero kernel. This contradiction gives $\Gamma=0$. Rank eight then proves the assertion. \end{proof} \begin{remark} The component hypothesis in Lemma~\ref{lem:m03-ten-space} holds for an integral full-span residue locus. It also holds for a conjugate pair whose components span distinct nonreal hyperplanes: a real linear form vanishing on one would be proportional both to that hyperplane equation and to its conjugate. Thus the lemma applies to that reduced union without assuming its coordinate ring is a domain. \end{remark}