\section{Quadratic poles with proper coefficient support} \label{sec:ps-proper-supports} Retain the notation of Section~\ref{sec:aq-two-poles}. For a simple real irreducible quadratic pole $Q$, let $U\subset\mathbb R^8$ be the total real span of its rank-two skew residue, of dimension $d$. For each geometric pole component, let $\mathcal A$ be the closed projective residue image, and write $\alpha=\dim\mathcal A$ and $s(a)=\operatorname{rank}B(a)$. Lemmas~\ref{lem:si-incidence} and~\ref{lem:cp-complex-ranks} give \[ 4\le s(a)\le\alpha\le5\quad(a\in\mathcal A),\qquad \mathcal A\subset V(a^THa),\qquad \operatorname{rank}(B(a)B(a)^T)=2s(a)-8. \] The image has no real points. Small supports $d\le5$ have already been excluded componentwise (Lemma~\ref{lem:si-small-support}). This section handles $d=6,7$, including rank-two real quadrics which split over $\mathbb C$. Write $G(a)=B(a)B(a)^T$. \subsection{Six-dimensional support} \begin{proposition}\label{prop:ps-six-structure} If $d=6$, then $H|_U$ is definite. For its positive choice of sign $h$, one has $G(a)=h(a)D$ on $U$, where $D\succeq0$ has rank eight and image $E$. The physical pole quadric is proportional to $Y^TDY$, has rank eight, and has kernel $K_X$. Two distinct quadratic poles cannot both have support dimension six. \end{proposition} \begin{proof} The restriction $H|_U$ is not zero, since a nondegenerate eight-form has no totally isotropic six-space. Its projective null locus in $\mathbb P(U_{\mathbb C})$ has dimension four, so each residue image fills a component. If the restricted rank is at least three, this quadric is irreducible. Any indefinite or degenerate real restriction would supply a real point. At ranks one or two every hyperplane component contains the nonzero real radical. Both alternatives are impossible. Thus $H|_U$ is definite of rank six. The image is its whole null quadric, with $\alpha=s=4$. Hence $G=0$ there, and quadratic divisibility gives $G=hD$. Evaluation at a real nonzero coefficient makes $D$ positive semidefinite of rank eight. All $B(a)$ for real $a\in U\setminus0$ have the same image $F=\operatorname{Im}D$. Radial columns put $M_XU\subset F\cap E$, so the lower projection of $F$ has dimension at most two. The lower skew pencil has exactly that fixed image at every nonzero $a$. If it had rank two, a six-parameter linear map into the one-dimensional space of skew maps on a fixed two-plane would have to have no nonzero kernel. This is impossible. Thus $F=E$. Contraction and polarization give \[ (P(Y)|_U)^T(P(Y)|_U)=(Y^TDY)H_0, \] where $H_0>0$ represents $h$. Modulo $Q$, the restricted pencil has corank at least two. Its Gram determinant therefore vanishes, so $Q\mid(Y^TDY)^6$. Equality of degrees gives proportionality of $Q$ and $Y^TDY$. Its rank and kernel follow from $\operatorname{Im}D=E$. For two six-spaces, choose a real nonzero vector in their intersection. Evaluating the two identities $G=h_iD_i$ there makes $D_1,D_2$ proportional, and therefore their quadrics proportional, contrary to distinctness. \end{proof} \begin{proposition}\label{prop:ps-six-exclusion} No generic sole-quadratic or two-quadratic branch has a component of support dimension six. \end{proposition} \begin{proof} For a whole simple quadratic, Proposition~\ref{prop:ps-six-structure} gives full-rank alignment of its intrinsic auxiliary family on the branch. Proposition~\ref{prop:aq-whole-aligned} excludes it. For two quadrics there is at most one six-support component. It has rank eight, hence is geometrically irreducible. Its support dimension is intrinsic: multiply the frame-free cross matrix by the pole factor, reduce at its generic point, and take the constant span of its image. This equals the coefficient support dimension in any eigenframe. On the generic factor cover, Galois automorphisms preserve this dimension and the factor degree. A nontrivial conjugate of the unique six-support factor would be the other quadratic and would also have support six, impossible on the branch. Thus this factor descends over $\mathbb R(X)$. Primitive descent and the shear argument of Section~\ref{sec:aq-two-poles} give a nonzero nonnegative aligned biquadratic family. Proposition~\ref{prop:aq-semidefinite-two} excludes it. This uses generic algebraic invariants on the factor cover, not preservation of real factorization at every specialization. \end{proof} \subsection{Seven-dimensional support with transpose rank four} \begin{lemma}\label{lem:ps-seven-rank-four} Suppose $d=7$ and the generic transpose rank on the residue image components is four. Then the quadratic entries of $G|_U$ span at most two scalar forms. For every real nonzero $a\in U$, $\operatorname{Im}B(a)=E$. Moreover $Q$ has rank eight and kernel $K_X$. \end{lemma} \begin{proof} Put $h=H|_U$, of rank at least six. On the residue image, $G=0$. First suppose $Q$ is geometrically irreducible. Its image is real, irreducible and spans $\mathbb P(U_{\mathbb C})=\mathbb P^6$. If $\alpha=5$ it is the whole $h$-quadric, and every entry of $G$ is proportional to $h$. If $\alpha=4$, its degree is at least three by the degree bound~\cite[p.~1, inequality~$(*)$]{EGHP2004}, and is even because a generic real complementary plane intersects it in conjugate pairs. An independent quadratic equation $f$ among the entries of $G$ puts it in the degree-four complete intersection $(h,f)$. Thus its degree is four and it exhausts that intersection, including the scheme structure: the complete intersection is unmixed and generically reduced, with no embedded components. All quadratic entries are in $\operatorname{span}\{h,f\}$. If $Q$ has rank two and both components have the same image, that image is real-defined and spans $\mathbb P(U_{\mathbb C})$, so the preceding paragraph applies. Otherwise take the conjugate image pair and let $d_1$ be the complex span dimension of one component. Dimension and the absence of real points exclude $d_1\le5$: the only possible image there is a projective linear space, whose intersection with its conjugate contains real vectors. If $d_1=6$, the image is the whole irreducible quadric in that hyperplane, of degree two; the conjugate hyperplanes are distinct. Their union is the reduced complete intersection $(h,\ell\bar\ell)$, giving the same two-form bound. If $d_1=7$, dimension five again gives the whole $h$-quadric. In dimension four, a real image uses the preceding parity argument; two distinct nondegenerate conjugate images have total degree at least six and cannot lie in a degree-four complete intersection. Consequently no independent quadratic $f$ exists and all entries are multiples of $h$. A positive semidefinite quadratic matrix of constant rank whose entries span at most two forms has fixed image. Indeed, on a real ball where one form has positive sign, the ratio of the other has an interval of values; the positive-semidefinite kernel-intersection identity and constant rank give equal kernels at two distinct parameters. Both coefficient matrices then have that fixed image. Apply this to $G$, of real rank eight. Its fixed image contains the radial seven-space $M_XU$, so its lower projection has dimension at most one. Skewness makes that projection zero, proving alignment. Thus $P(Y)|_U$ depends only on $v$. For every nonzero real $v$ it is injective: $B(a)$ maps onto $E$ for every nonzero real $a\in U$, so $B(a)^TM_Xv=P(v)a\ne0$. Every maximal seven-minor is divisible by $Q$ and some such minor is nonzero. Since that minor depends only on $v$, so does $Q$. A nonzero real vector in a smaller-rank semidefinite $Q(v)$ would contradict the injectivity. Hence $\operatorname{rank}Q=8$ and its full physical kernel is $K_X$. In particular $Q$ cannot have rank two. \end{proof} \begin{lemma}\label{lem:ps-seven-scalar} In the preceding situation $G|_U=h_0D_0$ for a positive definite quadratic $h_0$ and a fixed rank-eight positive semidefinite $D_0$. Moreover $(P|_U)^T(P|_U)=QH_1$ after a constant rescaling, with $H_1$ nondegenerate. Such a component cannot coexist with a distinct quadratic pole. \end{lemma} \begin{proof} Identify $E$ with the original orthonormal coefficient space. The square matrix $B_E(a)$ is invertible at real nonzero $a\in U$ and $B_E(a)g_X=-a$, with $|g_X|=1$. Hence $a^TG(a)^{-1}a=1$. If the scalar span is genuinely two-dimensional, then, since the pencil contains the positive definite $G_E(a_0)$ ($a_0\in U$ real nonzero), simultaneous congruence diagonalizes it as \[ G(a)=S\operatorname{diag}(f_j(a)I_{b_j})S^T, \qquad \sum b_j=8, \] with pairwise nonproportional quadratic forms $f_j$ in the span. Every $f_j$ is positive definite on the seven-space, hence irreducible, and the $f_j$ are pairwise nonassociate. The inverse identity is \[ \sum_j\frac{n_j(a)}{f_j(a)}=1, \qquad n_j=|\operatorname{pr}_jS^{-1}a|^2. \] Reduction modulo $f_j$ after clearing denominators gives $f_j\mid n_j$. If $n_j\ne0$, its rank is seven, whereas its sum-of-squares presentation has rank at most $b_j$. At least one is nonzero, so the only two-block possibility is sizes seven and one. But then \[ (\det B_E)^2=(\det S)^2 f^7g \] has an odd valuation, a contradiction. Thus the scalar span is one. Contraction gives $(P|_U)^T(P|_U)=(Y^TD_0Y)H_1$; its determinant modulo $Q$ makes $Q$ proportional to $Y^TD_0Y$. Modulo $Q$, the image of the eight-by-seven $P|_U$ is totally isotropic in the standard eight-space, hence has dimension at most four. Adding the last coefficient column gives $\operatorname{rank}P\le5$, so $Q^3\mid\det P$. A distinct quadratic pole $Q'$ requires $(Q')^2\mid\det P$ (Lemma~\ref{lem:cp-dvr-budget}). The divisor $Q^3(Q')^2$ has degree ten, contradicting $\deg\det P=8$. \end{proof} \begin{proposition}\label{prop:ps-two-seven} A two-quadratic branch has no component of support dimension seven. \end{proposition} \begin{proof} Write $\det P=cQ_1^2Q_2^2$. The local Smith exponents at either prime are $(1,1,0^6)$, and every seven-minor is divisible by both. Thus $A=\operatorname{adj}P/(cQ_1Q_2)$ is cubic and $AP=PA=Q_1Q_2I$. If the first residue lies in $U=\ker f$, then $fA=Q_1\ell(Y)^T$, with $\ell$ linear, because the image of $A$ modulo $Q_1$ is exactly the residue kernel plane. Multiplying by $P$ gives $P(Y)^T\ell(Y)=Q_2(Y)f^T$. Write $L:Y\mapsto\ell(Y)$. For real nonzero $a\in U$, put $P_aY=P(Y)a$. The last identity says $P_a^TL$ is skew; surjectivity of $P_a$ gives its kernel $\ker L$ and rank $m=\operatorname{rank}L$. Its image is the same $m$-space for every such $a$, and $m$ is even. For any nonzero real $g\in\operatorname{Im}L$, the vectors $N_ga$ with $a\in U$ lie in that space. Since $N_g$ is injective, it has dimension at least seven. Thus $m=8$ and $L$ is surjective. Over $\mathbb C$, $\operatorname{rank}(P_a^TL)=\operatorname{rank}P_a=s(a)$ is therefore even for every $a\in U_{\mathbb C}$. On each residue image the incidence bounds force $s=4$. Lemma~\ref{lem:ps-seven-scalar} now supplies the degree-ten contradiction. This argument also treats conjugate geometric factors of a real rank-two quadric. \end{proof} \subsection{Seven-dimensional support with transpose rank five} Here $\alpha=s=5$. Since $U$ is a hyperplane, $H|_U$ has rank at least six and its null quadric in $\mathbb P^6$ is irreducible. The residue image fills it. Absence of real points makes $H|_U$ definite of rank seven. Reverse the curvature sign if necessary and put $h=H|_U>0$; then $H$ has signature $(7,1)$. We first record an elementary seven-coefficient version of the fixed-image observation. If all images of $B(a)$, $a\in U$, lie in a fixed space $T$ of dimension at most nine, then they equal $E$. Indeed $M_XU\subset T\cap E$, so the lower projection $R$ of $T$ has dimension at most two. The lower skew pencil takes values in $\bigwedge^2R$, of dimension at most one, and hence vanishes at some nonzero real coefficient $a_0$. Its rank-eight image there is $E$, putting $E\subset T$. The lower projection of $T$ now has dimension at most one, so the entire lower skew pencil is zero. \begin{lemma}\label{lem:ps-seven-normal-form} In this rank-five case, either $B(U)$ has image $E$, or \begin{equation}\label{eq:ps-seven-balanced} G=hD+uu^T-vv^T \end{equation} with $D$ constant symmetric and $u,v$ real linear columns. The extraction includes divisorial rank-zero loci. \end{lemma} \begin{proof} The complex cone ring $S=\mathbb C[U]/(h)$ is factorial. For example write $h=x_1x_2+x_3^2+\cdots+x_7^2$; $x_1$ is prime and inverting it gives a Laurent polynomial ring, so Nagata's criterion \cite[Corollary to Theorem~6.3$'$, p.~18]{Samuel1964} applies. Alternatively, \cite[Theorem~8.2, p.~26]{Samuel1964} applies directly, since $h$ is nondegenerate in seven variables. At the generic point the transpose kernel is an isotropic three-space in the standard eight-form. Its binary quotient has determinant class $-1$ and is hyperbolic over the real function field. The two saturated isotropic line modules are rank-one reflexive and therefore principal over $S$. Choose homogeneous generators $p,r$. Then $G=t(pr^T+rp^T)$ over the fraction field. At every height-one prime, both generators are primitive and their symmetric cross product has a nonzero reduction. Thus $t$ has no negative valuation, and normality gives $t\in S$. Homogeneity gives $\deg t+\deg p+\deg r=2$. Each line is individually real-defined; conjugation changes its generator by a constant unit, removable by rescaling. Thus all generators and $t$ may be taken real. If both vector degrees are positive, they are one and $t$ is constant. Lifting and taking real sums and differences gives \eqref{eq:ps-seven-balanced}. Otherwise the lift is $G=hD+cb^T+bc^T$ with $c$ constant and $b$ quadratic. Compression to $c^\perp$ has the fixed form $hD_0$. Positivity of the full matrix puts its cross block in $\operatorname{Im}D_0$, so all images lie in a fixed space of dimension at most nine. The preceding observation gives alignment. The scalar $t$ retains all common divisorial factors, so no rank-zero divisor has been discarded. \end{proof} \begin{lemma}\label{lem:ps-seven-balanced-rank} For a nonaligned form \eqref{eq:ps-seven-balanced}, $D\succeq0$ has rank exactly eight. A constant hyperbolic change of the two columns gives \begin{equation}\label{eq:ps-seven-reduced} G=hD-ww^T+zz^T,\qquad w\in F:=\operatorname{Im}D,\qquad w^TD^+w=h, \end{equation} where the projection of $z$ to $F^\perp$ has rank at least two. \end{lemma} \begin{proof} For any fixed real physical vector $b$, choose nonzero $a\in U$ with $b^Tu(a)=b^Tv(a)=0$. Positivity gives $b^TDb\ge0$. Put $r=\operatorname{rank}D$ and $G_+=hD+uu^T$. Positivity of $G=G_+-vv^T$ gives $v\in\operatorname{Im}G_+$, and rank eight gives $\operatorname{rank}G_+\in\{8,9\}$. Therefore $7\le r\le9$. If $r=9$, both $u,v$ lie in $F$ and the fixed-nine-space observation aligns the image. Suppose $r=7$. The quotient map $\bar u:U\to V/F$ is injective, and $\operatorname{Im}B(a)=F+\mathbb Ru(a)$ for all real $a\ne0$. The radial vector $M_Xa$ belongs to this space. Two linear maps which are pointwise proportional, one of rank at least two, have constant proportionality; compare two independent images and their sums. Thus $\overline{M_Xa}=c\bar u(a)$ for a constant $c$. If $c=0$, $F=M_XU\subset E$; each lower image has dimension at most one and skewness kills it. This would put the injective $\bar u(U)$ in the one-dimensional quotient $E/F$, impossible. If $c\ne0$, the lower skew pencil has one fixed image $R=K_X^TF$ at all nonzero real coefficients. Its even rank is $0,2,4$, or $6$. Ranks two and four are impossible because their skew-operator spaces have dimensions one and six, less than seven. Rank six would give a cubic Pfaffian nowhere zero on the real unit sphere, impossible for an odd function on a connected sphere. Rank zero puts $F\subset E$, again contradicting the rank of the quotient radial map. Thus $r\ne7$. We have $r=8$. If $\operatorname{rank}\bar u\le1$, all images lie in a fixed nine-space and align. Otherwise positivity gives $\bar v=c\bar u$ with constant $c$, so $v=cu+w_0$ with $w_0\in F$. At points where $\bar u\ne0$, the rank drop from nine to eight is \[ c^2+h^{-1}w_0^TD^+w_0=1. \] This polynomial identity extends everywhere. Since $h>0$ on $U\setminus0$ and $D^+\succeq0$ on $F$, we have $c^2\le1$. If $|c|=1$, then $w_0=0$ and the image is fixed and aligned. Otherwise set $d=1-c^2>0$, $w=w_0/\sqrt d$, $z=\sqrt d\,u-cw$. Expansion gives \eqref{eq:ps-seven-reduced}; its quotient column retains rank at least two. \end{proof} \begin{proposition}\label{prop:ps-seven-spectral} A nonaligned rank-five seven-support branch forces a positive nonzero Jacobi root of multiplicity at least five. \end{proposition} \begin{proof} Project $B$ to $F^\perp$ in \eqref{eq:ps-seven-reduced}, writing the projection as $T$ and that of $z$ as $b$. Then $TT^T=bb^T$. Every row of $T$ with squared norm $b_i^2$ vanishes on the real hyperplane $b_i=0$, hence is $b_ie_i^T$ for a constant unit vector $e_i$. Cross-row identities give $e_i^Te_j=1$, so all are the same unit $e$. Thus $T=be^T$. The mixed block then gives $B(a)e=z(a)$. In orthonormal input coordinates beginning with $e$, write $B=[z,B_0]$, with $B_0$ valued in $F$ and \[ B_0B_0^T=hD-ww^T. \] Choose $S:\mathbb R^8\to F$ with $SS^T=D$, put $A_0=S^{-1}B_0$, $\eta_0=S^{-1}w$, and use $|\eta_0|^2=h$. The seven nonzero eigenvalues of $A_0A_0^T=hI-\eta_0\eta_0^T$ are $h$, so $A_0^TA_0=hI_7$. For $g\in e^\perp$, use $B(a)g=N_ga$ and an $h$-orthonormal basis of $U$. The resulting eight-by-seven matrix $A(g)$, linear in the seven variables $g$, satisfies \[ A(g)^TA(g)=q(g)I_7,\qquad q(g)=|g|^2. \] At the prime $(q)$ its image is isotropic in the standard eight-space, of dimension at most four. Each maximal seven-minor is therefore divisible by $q^3$. The signed cofactor vector has the form $q^3\eta(g)$, with $\eta$ linear. Cauchy--Binet gives $|\eta|^2=q$, and the normal-vector identity gives \[ A(g)A(g)^T=q(g)I_8-\eta(g)\eta(g)^T. \] Complete the $h$-orthonormal basis of $U$ by its $H$-orthogonal negative unit vector. In these coefficient coordinates $H=\Lambda=\operatorname{diag}(I_7,-1)$, with the reciprocal congruence on the factors leaving the physical Jacobi family unchanged. Let $n(g)$ be the last column of $N_g$ in this basis and $p(g)=S\eta(g)$. Then \begin{equation}\label{eq:ps-two-negative-squares} \Jac_{K_Xg}=q(g)D-p(g)p(g)^T-n(g)n(g)^T, \qquad p(g)^TD^+p(g)=q(g). \end{equation} In $\operatorname{Frac}(\mathbb R[g]/(q))$, the left side is nilpotent. The determinant identity for $V=[p,n]$ makes the symmetric two-by-two matrix $V^TV$ nilpotent. It has the form $\left[\begin{smallmatrix}a&b\\b&-a\end{smallmatrix}\right]$ with $a^2+b^2=0$. The rank-seven quadric is geometrically integral over $\mathbb R$, so its function field does not contain a square root of $-1$. Hence $V^TV=0$ in that field, and quadratic divisibility gives, in particular, $p^Tp=\alpha q$ with $\alpha>0$. The image $W_0=\operatorname{Im}p$ is a seven-space in $F$. The two norm identities in \eqref{eq:ps-two-negative-squares} give $D^+|_{W_0}=\alpha^{-1}I$ as a quadratic compression. A symmetric matrix with scalar compression on a hyperplane has that eigenvalue on the intersection of the hyperplane with the orthogonal complement of its one cross-block vector. Thus $D$ has eigenvalue $\alpha$ on a space $L_\alpha$ of dimension at least six. If this dimension is at least seven, subtracting the two rank-one terms leaves multiplicity at least five. If it is six, choose nonzero $g$ in the kernel of the linear map $g\mapsto\operatorname{pr}_{L_\alpha}p(g)$ from seven variables to six. The vectors of $L_\alpha$ orthogonal to $n(g)$ then give an $\alpha q(g)$-eigenspace of dimension at least five. Constancy of the real Osserman spectrum proves the assertion. \end{proof} \begin{proposition}\label{prop:ps-seven-sole} No remaining generic sole-quadratic branch has support dimension seven. \end{proposition} \begin{proof} For generic transpose rank four, Lemma~\ref{lem:ps-seven-rank-four} gives rank-eight alignment of the whole pole, and Proposition~\ref{prop:aq-whole-aligned} excludes it. For rank five, Lemma~\ref{lem:ps-seven-normal-form} either gives the same alignment or a balanced form. Any aligned alternative again has pole rank eight by the maximal-minor argument in Lemma~\ref{lem:ps-seven-rank-four}, which uses alignment only. The nonaligned balanced alternative has signature $(7,1)$ and Proposition~\ref{prop:ps-seven-spectral} forces a positive root of multiplicity at least five. Among the eleven remaining patterns with a reduced root of multiplicity seven (Section~\ref{sec:assembly}), only $(7,5,3)$ has a nonzero block of size five; it has no simple nonzero root and therefore cannot have negative index one. Every other applicable placement has positive multiplicities at most four. This contradiction finishes the case. \end{proof} Together the propositions exclude every proper support remaining after the small-support reduction. The whole-pole criterion was used only for a sole quadratic. The two-quadratic conclusions instead used componentwise composition, algebraic descent, and the tight degree-eight determinant. Alignment on a proper coefficient space was never substituted directly for the full-coefficient closed-kernel hypothesis.