\section{Small supports and the residue-image incidence} \label{sec:si-support-incidence} Use the normalization of Section~\ref{sec:coherent-transpose}. The shifted Jacobi rank is $\nu\in\{7,8\}$, the lower dimension is $k=16-\nu$, and the actual lower pencil $P(Y):\R^\nu\to\R^k$ is linear and generically injective for $Y\in X^\perp$. Its extension to $\R^{16}$ has $P(X)=0$. The maps $N_g$ and $B(a)g=N_ga$ are as in \eqref{eq:cp-N}. In particular, Lemma~\ref{lem:cp-L20} gives \begin{equation}\label{eq:si-fixed-output} P(Y)U\subset T\text{ for all }Y,\quad U\ne0 \quad\Longrightarrow\quad T=\R^k \end{equation} for fixed real coefficient and output subspaces $U,T$. This identity is valid through exceptional partner directions by polynomial continuation; no limiting rational frame is required. \subsection{Four- and five-dimensional quadratic supports} For a specified simple irreducible real quadratic pole $Q$, let $W$ be its nonzero rank-two skew residue over the fraction field of $\R[Y]/(Q)$. Its \emph{real support} is the smallest constant real coefficient subspace whose complexified or residue-field extension contains $\operatorname{im}W$. This includes both geometric components when $Q$ splits over $\mathbb C$. \begin{lemma}\label{lem:si-small-support} A simple quadratic component with rank-two residue cannot have real support of dimension at most five. This is componentwise: other polar factors may be present. \end{lemma} \begin{proof} Suppose the support is contained in a fixed real $d$-space $U$. At the prime $(Q)$, the kernel of $P|_U$ contains the rank-two residue image. Rectangular Smith form therefore gives at least two positive valuations. Every maximal $d$-minor of $P|_U$ is divisible by $Q^2$. Some such minor is nonzero by generic injectivity, and has degree $d$; thus $d\geq4$. If $d=4$, all maximal minors are constant multiples of $Q^2$. Their nonzero Pl\"ucker vector has constant projective direction, so the generic images $P(Y)U$ are one fixed real four-space. Projection to its complement is polynomial and vanishes generically, hence for every $Y$. This contradicts \eqref{eq:si-fixed-output}, since $k\geq8$. If $d=5$, the maximal minors give \begin{equation}\label{eq:si-linear-plucker} \bigwedge\nolimits^5(P(Y)|_U)=Q(Y)^2\ell(Y) \end{equation} with $\ell$ a nonzero real homogeneous linear map into $\bigwedge^5\R^k$. Its generic values are decomposable. The Pl\"ucker relations are polynomial, so every value is decomposable. The resulting five-spaces have one of three arrangements: constant; all contain a fixed four-space $T_0$; or all lie in a fixed six-space $T_1$. Here is the elementary linear-Grassmannian argument. Two independent decomposable five-vectors whose sum is decomposable determine five-spaces meeting in dimension four. Fix two distinct members, with intersection $T_0$ and sum $T_1$. A third member adjacent to both either contains $T_0$ or lies in $T_1$. If one member contains $T_0$ but is not contained in $T_1$, a member in $T_1$ not containing $T_0$ would meet it in dimension at most three, a contradiction. Thus the family is a star through $T_0$ or a top inside $T_1$; the one-dimensional Pl\"ucker image is the constant case. Constant output five and a top inside output six both contradict \eqref{eq:si-fixed-output}. In the star case project $P|_U$ modulo $T_0$. The resulting linear matrix family has rank at most one, first generically and then everywhere by its two-minors. A linear space of rank-one matrices has either a common image line or a common coefficient kernel of codimension one. Indeed, a sum of two rank-one tensors with independent left and right factors has rank two; applying this to all pairs gives the dichotomy. If $u_1\otimes v_1,u_2\otimes v_2$ have parallel left factors but independent right factors, any third member $u_3\otimes v_3$ with $u_3$ independent of $u_1$ would need $v_3$ parallel to both $v_1$ and $v_2$, which is impossible. A common image line confines $P(Y)U$ to a fixed five-space. A common kernel is a fixed four-space $U_0\subset U$ with $P(Y)U_0\subset T_0$ for all $Y$. Both violate \eqref{eq:si-fixed-output}. Only divisibility by the specified factor $Q^2$ was used. No sole-denominator or minimal-support refinement enters the star case. \end{proof} \begin{proposition}[Isolated skew corrections]\label{prop:si-isolated-skew} Let $U\subset\R^\nu$ be a fixed real coefficient space of dimension at most five, and write $P_U=P|_U$ in a fixed orthonormal basis of $U$. If $\Gamma(Y)$ is a rational skew-symmetric endomorphism of $U$ and $P_U(Y)\Gamma(Y)$ is polynomial, then $\Gamma$ has no pole along an irreducible real quadratic. No assertion that $\Gamma$ is the full actual residue or correction is required. \end{proposition} \begin{proof} If $\Gamma$ had a pole of any order $e>0$ at a quadratic $Q$, its leading coefficient $W=(Q^e\Gamma)\bmod Q$ would be nonzero and skew, hence of rank at least two. Polynomiality of $P_U\Gamma$ gives $P_UW=0$ modulo $Q$. Thus $P_U$ has corank at least two there, and every maximal minor is divisible by $Q^2$. The degree count, constant-output case, and linear Pl\"ucker star/top argument in the proof of Lemma~\ref{lem:si-small-support} now apply without change. Each gives a nonzero fixed real coefficient subspace mapped by the \emph{actual} pencil $P$ into a proper fixed output, contradicting \eqref{eq:si-fixed-output}. Neither $H$-isotropy nor a simple-pole assumption is used in this proposition; the essential inputs are skewness, generic injectivity of $P_U$, and the actual output property \eqref{eq:si-fixed-output} of $P$. \end{proof} \subsection{A real-irreducible quartic has full real support} \begin{lemma}\label{lem:si-quartic-support} In the case $\nu=8$, a genuine simple real-irreducible quartic pole has real residue support of dimension eight, including when it splits into two conjugate complex quadratics. \end{lemma} \begin{proof} Let the quartic be $f$. The local budget of Lemma~\ref{lem:cp-dvr-budget} and $\deg\det P=8$ give \[ \det P=c f^2,\qquad c\in\R\setminus\{0\}. \] The nonzero simple skew residue has positive even rank. The same budget forces rank two and Smith exponents $(1,1,0^6)$ at $(f)$. Every seven-minor is divisible by $f$, so \[ A=\frac{\operatorname{adj}P}{cf} \] is a homogeneous cubic polynomial matrix with $AP=PA=fI$. Modulo $f$, its image is the two-dimensional kernel of $P$ and therefore equals the image of the rank-two residue $W$. If a constant real row $b^T$ annihilates the real support of $W$, it annihilates $A$ modulo $f$. Each entry of $b^TA$ has degree three, less than $\deg f=4$, so $b^TA=0$ identically. Multiplication by $P$ gives $fb^T=0$, hence $b=0$. The argument takes place over the irreducible real prime $(f)$. It makes no claim that either individual complex component has full complex support. \end{proof} \subsection{Dimension and rank of the closed image} From now on let $\nu=k=8$, and let $H=\Lambda^{-1}$ be indefinite. Put $h(a)=a^THa$ and $G(a)=B(a)B(a)^T$. Fix one irreducible \emph{complex} pole hypersurface $Z\subset\mathbb P(X^\perp_{\mathbb C}) =\mathbb P^{14}$ with generic residue rank two. Let $I$ be the closure of its projectivized residue-image bundle in $Z\times\mathbb P^7$, and let $A$ be the closed image of $I$ in $\mathbb P^7$. These are irreducible projective varieties, and $\dim I=13+1=14$. Write $\alpha=\dim A$ and let $s$ be the generic rank of $B(a)$ on $A$. \begin{lemma}\label{lem:si-incidence} One has \[ 4\leq s\leq\alpha\leq5. \] Moreover $A$ has no real point, $A\subset V(h)$, and $\rank G(a)=2\rank B(a)-8$ for every nonzero complex $a$. \end{lemma} \begin{proof} The rank identity and the lower bound four are Lemma~\ref{lem:cp-complex-ranks}. The residue identities \eqref{eq:cp-residue} imply $A\subset V(h)$, so initially $\alpha\leq6$. They also put every fibre of $I\to A$ inside \[ L_a=\{[Y]\in\mathbb P^{14}:P(Y)a=0\}. \] The linear map $Y\mapsto P(Y)a$ has rank $\rank B(a)$, since its transpose is $B(a)$ and $P(X)=0$. Generically, $\dim L_a=14-s$. The generic incidence fibre has dimension $14-\alpha$, whence $s\leq\alpha$. The minors giving $\rank B\leq\alpha$ vanish generically on $A$ and therefore everywhere on $A$. A nonzero real coefficient has rank eight by Lemma~\ref{lem:cp-L20}, so $A$ has no real point. If $\alpha=6$, it would equal the irreducible six-dimensional quadric $V(h)$, which has real points because $h$ is indefinite. Hence $\alpha\leq5$. \end{proof} \subsection{Eliminating the unequal full-support type} \begin{proposition}\label{prop:si-five-four} Let a genuine simple irreducible real quadratic or quartic pole have total real residue support eight. If a complex residue-image component has $\alpha=5$ and $s=4$, then the full transpose image is $E_X$ at this anchor: $\operatorname{Im}B(a)=E_X$ for every real $a\ne0$. \end{proposition} \begin{proof} The rank identity makes $G=0$ on that closed image $A$. If the real pole polynomial is geometrically irreducible, conjugation preserves its unique complex component, so $A$ is real-defined. Its span is then a real-defined subspace containing the residue image, hence equals $\mathbb P^7$ by total support eight. If $A$ is real-defined and spans $\mathbb P^7$, the bound $\deg A\ge1+\operatorname{codim}A$ for nondegenerate varieties \cite[p.~1, $(*)$]{EGHP2004} gives $\deg A\geq3$. Since $A$ has no real points, intersection with a generic transverse real complementary plane makes its degree even, hence at least four. If every entry of $G$ is a multiple of $h$, evaluation at a nonzero real $h$-null vector contradicts rank eight. Otherwise choose a real entry $f$ independent of $h$. The degree-four complete intersection $V(h,f)$ contains $A$. The complete intersection $(h,f)$ is Cohen--Macaulay, hence unmixed. Since $\deg V(h,f)=4\le\deg A$, the variety $A$ is its only component, of multiplicity one. Thus $(h,f)$ is generically reduced and unmixed, hence radical, and its degree-two part is $\operatorname{span}(h,f)$. The scalar-span criterion, Lemma~\ref{lem:scalar-span-fixed-image} and Remark~\ref{rem:scalar-span-radial}, gives the claimed image. An irreducible real pole polynomial is geometrically irreducible or has exactly two conjugate irreducible factors. In the latter case write $A,\overline A$ for the two images and $U,\overline U$ for their vector spans. Total real support eight means $U+\overline U=\mathbb C^8$. Both images have dimension five and $G$ vanishes on both. Put $d=\dim_{\mathbb C}U$. If $d\leq6$, dimension forces $d=6$ and $A=\mathbb P(U)$. The conjugation-invariant intersection $U\cap\overline U$ has dimension at least four and contains a nonzero real vector, giving a forbidden real point of $A$. If $d=7$, the restriction $h|_U$ has rank at least six and its irreducible quadric fivefold equals $A$. The two spans are distinct hyperplanes, say $\ell=0$ and $\overline\ell=0$. Their two degree-two quadrics exhaust the degree-four complete intersection $V(h,\ell\overline\ell)$, which is their reduced union by the same unmixedness argument. Its quadratic ideal has scalar span two, so the criterion again applies. If $d=8$ and $A=\overline A$, use the first paragraph. If $d=8$ and they are distinct, each has degree at least three \cite[p.~1, $(*)$]{EGHP2004}. Any quadratic entry $f$ independent of $h$ would put their union of degree at least six inside $V(h,f)$ of degree four, impossible. Thus all entries would be multiples of $h$, already contradicted. These alternatives also cover a real quartic splitting into conjugate quadratics; no individual full-span assumption was made in the split cases. \end{proof} The proposition gives a pointwise full-image identity. On a nonempty real open branch of anchors, the argument following Remark~\ref{rem:m03-scope} (see the proof in Section~\ref{sec:assembly}) and Lemma~\ref{lem:m03-closed-kernel} turn it into the classification conclusion. It is enough here to use the earlier scalar-span criterion; the stronger two-form radial-Gram theorem is not needed for this incidence reduction. \subsection{Saturation is conditional on equal dimension and rank} \begin{lemma}\label{lem:si-saturated-fibres} For the incidence above, suppose $\alpha=s$. For generic $a\in A$ the fibre $I_a$ equals the entire projective linear space $L_a$. Consequently two distinct complex pole hypersurfaces with this equality cannot have the same residue image. \end{lemma} \begin{proof} The closed fibre $I_a$ lies in $L_a$ and has dimension $14-\alpha=14-s=\dim L_a$ at a generic point. A proper closed subset of an irreducible projective linear space has smaller dimension, so $I_a=L_a$. If two pole incidences had the same image $A$ and satisfied this equality, their fibres would coincide over one common dense open subset of $A$. Their irreducible closures would therefore coincide. Projecting those closures to $\mathbb P^{14}$ would give the same pole hypersurface, contrary to distinctness. No reality or full-span assumption on an individual image enters this lemma. \end{proof} \begin{remark}\label{rem:si-order} The order of use is essential. First remove proper real supports, using the small-support lemma above and Propositions~\ref{prop:ps-six-exclusion}, \ref{prop:ps-seven-sole} and~\ref{prop:ps-two-seven}, with Lemma~\ref{lem:si-quartic-support} for quartics. Then every remaining real component has support eight. Lemma~\ref{lem:si-incidence} gives the possibilities $(4,4)$, $(5,4)$, $(5,5)$; Proposition~\ref{prop:si-five-four} reduces the unequal one to full transpose alignment, which ends in the Clifford conclusion on an open branch by Lemma~\ref{lem:m03-closed-kernel}. Only at this point is $\alpha=s$ available, and Lemma~\ref{lem:si-saturated-fibres} gives saturation and distinct images. The saturation lemma does not itself prove the equality. \end{remark}