\section{Exclusion of a double quadratic pole} \label{sec:dq-exclusion} We work in the indefinite rank-eight case. Retain the coherent factors, the real quadratic-pole condition, and the intrinsic denominator from Sections~\ref{sec:coherent-transpose} and~\ref{sec:real-pole-types}. At a regular real unit anchor $X$, choose orthonormal frames $M_X$ of $E_X=\operatorname{Im}\Jac_X$ and $K_X$ of $\ker\Jac_X$. Put $H=\Lambda^{-1}$ and write, for $Y\perp X$, \[ M_Y=M_XF(Y)+K_XP(Y),\qquad F=F_0+\Omega H, \qquad \Omega^T=-\Omega. \] Here $F_0,P$ are linear, $P$ is generically invertible, and $P\Omega$ is polynomial. The latter identity follows from the polynomial lower--upper Jacobi block $P\Lambda F^T$. The original lower frame $K_X$ will remain orthonormal throughout. \begin{proposition}\label{prop:dq-exclusion} The reduced common denominator cannot be $Q_X^2$ on a nonempty real open branch of regular anchors, where $Q_X$ is an irreducible real positive semidefinite quadratic. \end{proposition} \subsection{A degree bound for an entire skew block} \begin{lemma}\label{lem:dq-block-degree} Let $P_0$ be a real polynomial $m\times b$ matrix with linear entries and generic column rank $b$. Let $\Omega_0$ be a skew rational $b\times b$ matrix such that $P_0\Omega_0$ is polynomial. If $\Omega_0$ has pole order $e$ at a real irreducible quadratic $Q$, then every maximal $b$-minor of $P_0$ is divisible by $Q^{2e}$. In particular, $b\ge4e$. \end{lemma} \begin{proof} Work in the discrete valuation ring at $Q$, and take a Smith form \[ P_0=U\begin{bmatrix}\operatorname{diag}(Q^{a_1},\ldots,Q^{a_b})\\0\end{bmatrix}V \] with $U,V$ invertible over that ring. Congruence gives the skew matrix $\widehat\Omega=V\Omega_0V^T$, still of pole order $e$. Choose an off-diagonal entry of valuation $-e$. Polynomiality of $P_0\Omega_0$, in its two corresponding rows, forces both associated Smith exponents to be at least $e$. Their sum is therefore at least $2e$, which is a lower bound for the valuation of every maximal minor. At least one maximal minor is nonzero; its polynomial degree is at most $b$, whereas $\deg Q^{2e}=4e$. \end{proof} Under the double-pole hypothesis this lemma applied to $P,\Omega$ gives \begin{equation}\label{eq:dq-determinant} \det P=c_XQ_X^4,\qquad c_X\ne0, \end{equation} by degree eight. Let $g_X$ represent $X$ in the original lower frame. The radial row is $g_X^TP(Y)=-Y^TM_X$. For $Y\in E_0(X)$, the reduced zero eigenspace, this row vanishes. Equation \eqref{eq:dq-determinant} implies $Q_X(Y)=0$ there. Since $Q_X$ is positive semidefinite, every such vector lies in its radical. Together with shear invariance, this shows that the quadratic pole form annihilates the full kernel of $\Jac_X$ and has rank at most eight. \subsection{The aligned auxiliary family} Let $D(X,Y)$ be the intrinsic reduced denominator, including pole orders. By Proposition~\ref{prop:rp-intrinsic-denominator} it has bidegree $(4,4)$ in the present branch. We justify the square root globally, rather than applying the degree-two intrinsic-divisor statement directly to $D$. Factor $D$ in $\R[X,Y]$. By the same proposition there is no pure-slot factor. Thus each factor is primitive in $Y$, and Gauss's lemma identifies its factorization with the one over $\R(X)[Y]$. On a suitable nonempty Zariski-open set of anchors, the fibre of $D$ is the actual reduced denominator (Proposition~\ref{prop:rp-intrinsic-denominator}), and the nonconstant fibres of distinct such factors remain pairwise coprime and squarefree: distinct irreducible factors are coprime over $\R(X)$, and squarefree over its algebraic closure because the characteristic is zero; after a generic linear change of the $Y$-coordinates, their resultants and discriminants with respect to one coordinate are nonzero polynomials, and they stay nonzero off a proper algebraic set of anchors. This set meets the real open branch. At such an anchor, each factor fibre divides $Q_X^2$ and is squarefree, hence is proportional to the irreducible real quadratic $Q_X$. Pairwise coprimeness permits only one factor, and its exponent must be two. There is therefore exactly one global irreducible factor, with multiplicity two. Consequently \begin{equation}\label{eq:dq-global-square} D=c\,\mathcal Q(X,Y)^2, \end{equation} where $\mathcal Q$ is a real polynomial of bidegree $(2,2)$ and $c$ is a nonzero constant. No anchor-dependent square root has been chosen. Both shear identities for $D$ pass to $\mathcal Q$. Indeed their squares agree in the polynomial ring including the shear parameter; the integral-domain property gives a common sign, and parameter zero fixes that sign positively. The elementary shears generate the exchange $(X,Y)\mapsto(Y,-X)$, and the even bidegrees give \[ \mathcal Q(Y,X)=\mathcal Q(X,Y),\qquad \mathcal Q(X,Y+tX)=\mathcal Q(X,Y),\qquad \mathcal Q(X,X)=0. \] Write $\mathcal Q(X,Y)=\langle A_XY,Y\rangle$. Then $A_X$ is a symmetric quadratic family and $A_XX=0$. By \eqref{eq:reconstruct} it is the Jacobi family of an algebraic curvature tensor; this is not used below. Anchors at which Proposition~\ref{prop:rp-denominator-types} applies and the fibre of $D$ is $D_X$ are dense. At each of them $c\,\mathcal Q(X,\cdot)^2=D_X$ is one of the listed types; a square can only be $Q^2$, so $\mathcal Q(X,\cdot)$ is an irreducible positive semidefinite quadratic with a genuine double pole, and the radial-row argument above applies at $X$. Semidefiniteness persists everywhere by continuity. Choose a nonzero fibre positive, at $X_0$. Symmetry gives $\mathcal Q(Y,X_0)=\mathcal Q(X_0,Y)>0$ outside its kernel, so $A_Y$ is positive semidefinite on that dense set, and then everywhere. The radial-row argument above gives \begin{equation}\label{eq:dq-aligned-family} \ker\Jac_X\subset\ker A_X,\qquad A_X\succeq0. \end{equation} The inclusion extends from this dense set of anchors by continuity of the kernel projector. No assertion that $A$ is Osserman or commutes with $\Jac$ is used. \begin{lemma}\label{lem:dq-common-coefficient} For the fixed anchor $X$ of the branch and, as a rational identity, for $Y\in X^\perp$, there is a constant real symmetric positive semidefinite matrix $C$ such that \[ A_{sX+Y}=(sM_X+M_Y)C(sM_X+M_Y)^T. \] Moreover $\rank C=\rank Q_X$ and \begin{equation}\label{eq:dq-commutator} \Omega HC-CH\Omega\quad\text{is polynomial}. \end{equation} \end{lemma} \begin{proof} For a generic partner $Y$, the matrix $B=[M_X,M_Y]$ has full column rank sixteen. Image inclusion in \eqref{eq:dq-aligned-family} gives $A_{sX+tY}=BZ(s,t)B^T$ with $Z$ a symmetric quadratic polynomial block matrix. Its values at the axes force the form \[ Z=\begin{bmatrix}s^2C+stE&stF_1\\stF_1^T&t^2C_1+stH_1\end{bmatrix}. \] For generic real $s,t$, its image lies in the image of $[sI,tI]^T$. Hence $[-tI,sI]Z=0$ identically. Coefficient comparison gives $E=H_1=0$ and $F_1=C_1=C$. Since $C=M_X^TA_XM_X$, it is independent of $Y$ and positive semidefinite. The fibre defining $A_X$ is proportional to $Q_X$, giving the rank. The mixed coefficient of $A_{sX+Y}$ makes $FC+CF^T$ polynomial and linear in $Y$. Substitution of $F=F_0+\Omega H$ gives \eqref{eq:dq-commutator}. \end{proof} \subsection{The whole polar correction and total isotropy} Consider the constant real linear map on skew matrices $\mathcal L(Z)=ZHC-CHZ$. Choose a constant complement to its kernel. On that complement $\mathcal L$ is injective, so \eqref{eq:dq-commutator} shows that the complementary part of $\Omega$ is polynomial. Subtract it. The remaining skew correction, still denoted by $\Omega$, has the same genuine double pole and satisfies \begin{equation}\label{eq:dq-exact-kernel} \Omega HC=CH\Omega,\qquad P\Omega\text{ polynomial}. \end{equation} This is an identity for the \emph{entire rational correction}, not only its highest residue. \begin{proof}[Completion of Proposition~\ref{prop:dq-exclusion}] Let $r=\rank C=\rank Q_X$ and $s=8-r$. Under a constant coefficient change $G$, the simultaneous transformations are \[ C'=G^{-1}CG^{-T},\quad H'=G^THG,\quad \Omega'=G^{-1}\Omega G^{-T},\quad P'=PG. \] They preserve skewness, the exact identities \eqref{eq:dq-exact-kernel}, and generic injectivity of column subpencils. They do not change the lower output frame. First normalize $C=\operatorname{diag}(I_r,0_s)$. Let $k$ be the nullity of the restriction of $H$ to the first $r$-space and set $d=r-k$, $\ell=s-k$. Nondegeneracy of $H$ gives $k\le s$. Further $C$-preserving changes put the two forms, on spaces of dimensions $d,k,k,\ell$, in the form \begin{equation}\label{eq:dq-form-normalization} C=\operatorname{diag}(I_d,I_k,0_k,0_\ell),\qquad H=\begin{bmatrix} H_0&0&0&0\\0&0&I_k&0\\0&I_k&0&0\\0&0&0&H_1 \end{bmatrix}, \end{equation} where $H_0,H_1$ are nondegenerate and empty blocks are allowed. Here is an explicit justification. Orthogonally split the first $r$-space into its nondegenerate part and radical. The radical's pairing with the last $s$-space has row rank $k$, or $H$ would be singular. Changes of shape $\left[\begin{smallmatrix}O&T\\0&N\end{smallmatrix}\right]$, with $O$ orthogonal and $N$ invertible, preserve $C$. The first $d$ rows of $T$ remove the cross terms from $H_0$; $N$ normalizes the radical pairing to $[I_k,0]$. The remaining rows of $T$ remove the diagonal and off-diagonal terms on the paired $k$-space. This gives \eqref{eq:dq-form-normalization}; nondegeneracy gives that of $H_1$. Direct comparison in $\Omega HC=CH\Omega$ now gives the exact rational block form \begin{equation}\label{eq:dq-whole-blocks} \Omega=\begin{bmatrix} Z&0&0&0\\ 0&B_0&M&T_0\\ 0&M&0&0\\ 0&-T_0^T&0&V_0 \end{bmatrix},\qquad ZH_0=H_0Z, \end{equation} where $Z,B_0,M,V_0$ are skew. In particular, both displayed off-diagonal copies of $M$ have the same sign because $M^T=-M$. Diagonalize $H_0$ orthogonally. Its commuting block $Z$ is a direct sum of skew blocks on the nonzero eigenspaces of $H_0$. Each such space is definite for the full indefinite form $H$, and therefore has dimension at most seven. The remaining skew block, on the last three spaces of \eqref{eq:dq-whole-blocks}, has dimension $s+k=8-d$. If $d>0$, it also has dimension at most seven. Because these are exact constant block decompositions of the entire correction, the corresponding columns of $P\Omega$ are separately $P_{\mathrm{block}}\Omega_{\mathrm{block}}$ and are polynomial. Each $P_{\mathrm{block}}$ is linear and generically injective. Lemma~\ref{lem:dq-block-degree} excludes an order-two quadratic pole in every block of dimension at most seven. If $d>0$, no block could carry the genuine double pole. Thus $d=0$, or equivalently \begin{equation}\label{eq:dq-isotropic-range} CHC=0. \end{equation} The range of $C$ is a real totally $H$-isotropic space, so $r\le\operatorname{Witt}(H)\le4$. This excludes all ranks five through eight without an anisotropy argument over a residue field. \emph{Ranks two and three.} The quadratic $Q_X$ is real irreducible and semidefinite, so $r\ge2$. With $d=0$, we have $k=r$, $\ell=8-2r$, and \[ \Omega=\begin{bmatrix}B_0&M&T_0\\M&0&0\\-T_0^T&0&V_0\end{bmatrix}. \] Write $P=[P_R,P_{R'},P_L]$ for the column blocks on the three spaces. The columns indexed by the middle $r$-space in $P\Omega$ are exactly $P_R M$. For $r=2,3$, even a simple pole of the skew block $M$ would require $r\ge4$ by Lemma~\ref{lem:dq-block-degree}. All its possible denominators divide $Q_X^2$, so $M$ is polynomial. Subtract its polynomial skew off-diagonal block. The remaining entire polar correction is supported on the first and last spaces, of combined dimension $8-r<8$. It still has a genuine double pole, contradicting the same degree lemma. Hence ranks two and three are impossible. \emph{The last rank-four case.} Now $r=4$, $H$ has signature $(4,4)$, and the constant coefficient normalization gives \[ H'=\Lambda'=\begin{bmatrix}0&I_4\\I_4&0\end{bmatrix},\quad C'=\begin{bmatrix}I_4&0\\0&0\end{bmatrix},\quad \Omega'=\begin{bmatrix}B_0&M\\M&0\end{bmatrix},\quad P'=[P_R,P_N]. \] The four-column identity $P_RM$ polynomial excludes a double pole of $M$. If $M$ were polynomial, subtracting it would leave a double-polar skew block supported on the first four-space, also impossible. Thus $M$ has a genuine simple $Q_X$-pole. Every maximal four-minor of $P_R$ is consequently divisible by $Q_X^2$. Its degree is four, so all those minors are constant multiples of $Q_X^2$, with at least one nonzero. Their Pl\"ucker coordinates describe one fixed real four-dimensional output space $T$. Thus $\operatorname{Im}P_R(Y)\subset T$ on the generic locus, and then for every $Y$ by polynomial continuation. Choose a nonzero vector $\eta\in T^\perp$ in the \emph{original} Euclidean lower output coordinates, and put $z=K_X\eta\ne0$. Since only coefficient coordinates were changed, \[ \eta^TP'(Y)=(0,b(Y)^T),\qquad P'\Lambda'P'^T=P\Lambda P^T. \] The hyperbolic matrix $\Lambda'$ therefore gives \[ \langle\Jac_{tX+Y}z,z\rangle =\eta^TP'(Y)\Lambda'P'(Y)^T\eta=0 \] for all real $t,Y$. These are polynomial identities, so rational frame singularities do not remove any directions. Sectional symmetry gives $\langle\Jac_zV,V\rangle=0$ for every real $V$. Symmetry and polarization imply $\Jac_z=0$, contradicting rank eight at the nonzero vector $z$. This is the sectional-symmetry fixed-output obstruction of the actual lower pencil. All possible ranks $2\le r\le8$ have been excluded. At no point was a nonorthogonal lower output change identified with a physical orthonormal frame, or a highest-residue block substituted for an entire rational polar block. The double-quadratic branch is impossible. \end{proof}