\providecommand{\AQisolatedref}{Proposition~\ref{prop:si-isolated-skew}} \section{Aligned quadratic families and two quadratic poles} \label{sec:aq-two-poles} We use the coherent construction at Jacobi rank eight, in the remaining nonscalar patterns. At a regular unit anchor $X$, let $E=\operatorname{Im}\Jac_X$, $K=\ker\Jac_X$, and use orthonormal frames $M_X,K_X$. Write $Y=M_Xv+u\in X^\perp$, with $u\in K\cap X^\perp$, and \[ M_Y=M_XF(Y)+K_XP(Y),\qquad H=\Lambda^{-1}. \] The lower pencil $P$ is linear and generically invertible; its radial row is $-v^T$. The transpose pencil will be denoted $B(a)$, so $B(a)^TY=P(Y)a$ and $B(a)g=N_ga$. It has rank eight for nonzero real $a$, satisfies $B(a)g_X=-M_Xa$, and $K_X^TB(a)$ is skew-symmetric. The earlier intrinsic-denominator construction is used here in its frame-free form. Its cross matrix $\mathcal T(X,Y)$ satisfies $\mathcal T+\mathcal T^T=2\mathcal B(X,Y)$, has its two images in $E_Y,E_X$, and obeys the two shear identities. Its reduced common denominator $D(X,Y)$ is bihomogeneous, has no pure-slot factor, and is invariant under both shears. Generic specialization has no common cancellation. A whole quadratic denominator is a nonnegative symmetric biquadratic sectional form, hence determines an auxiliary algebraic-curvature Jacobi family. These facts (Proposition~\ref{prop:rp-intrinsic-denominator}, where the cross matrix is written $T$), the generic-plane absence of real poles (Lemma~\ref{lem:rp-no-real-binary-poles}), and the componentwise exclusion of quadratic residue supports of dimension at most five (Lemma~\ref{lem:si-small-support}) are incoming results. All global branch arguments below take place on a nonempty real open anchor region, as in Lemma~\ref{lem:asm-partition}. \subsection{Common coefficients and a whole aligned quadratic} \begin{lemma}\label{lem:aq-common-coefficient} Let $A_z$ be a quadratic symmetric family with $\operatorname{Im}A_z\subset\operatorname{Im}\Jac_z$ for all real nonzero $z$. In the canonical coherent frame there is a constant symmetric matrix $C=M_X^TA_XM_X$ such that $A_z=M_zCM_z^T$. If $A_z\succeq0$, then $C\succeq0$. \end{lemma} \begin{proof} At a generic pair $X,Y$, the concatenation $\widehat M=[M_X,M_Y]$ is injective. A constant left inverse expresses the quadratic family $A_{sX+tY}$ uniquely as $\widehat MZ(s,t)\widehat M^T$, with $Z$ symmetric quadratic. Its values at the coordinate axes give \[ Z(s,t)=\begin{bmatrix} s^2C+stZ_{11}&stZ_{12}\\stZ_{12}^T&t^2C_1+stZ_{22} \end{bmatrix}. \] Image inclusion implies $[-tI,sI]Z=0$. Comparing its coefficients gives $Z_{11}=Z_{22}=0$ and $Z_{12}=C_1=C$. This proves the identity on the generic pair locus and then rationally everywhere. The coefficient is fixed at $X$, and positivity there proves the last assertion. \end{proof} We record the anchor-change step used for a whole quadratic. The coherent-pair construction supplies a nonempty open set of disjoint partners at each fixed nonzero real anchor: the proof of \cite[Lemma~2(1), arXiv:math/0204258v2, p.~7]{Nik2005} fixes an arbitrary unit $X$ before proving openness and density in $Y$. The intrinsic matrix $\mathcal T$ is regular near such a pair. \begin{lemma}\label{lem:aq-affine-anchor} Suppose the tensor is not Clifford and its whole intrinsic denominator is quadratic in each slot. Fix a regular unit $X$ and put $Q(Y)=D(X,Y)$. Suppose $Q$ is unchanged by translations in a real subspace $U\subset X^\perp$, and the full coherent factor is affine in those directions. If $A$ is the auxiliary family, then \[ A_V=c(V)A_X\quad(V\in\mathbb RX+U), \] where $c$ is a positive definite quadratic form and $c(X)=1$. \end{lemma} \begin{proof} First no nonzero real anchor $V$ has $D(V,Y)\equiv0$. Indeed, nonnegativity in the first slot would give $D(V+tZ,Y)=t^2D(Z,Y)$. Put $N=D\mathcal T$, a polynomial. At a disjoint pair $(V,Y)$, the coefficient $N_2(Z,Y)$ of $t^2$ in $N(V+tZ,Y)$ is $D(Z,Y)\mathcal T(V,Y)$. Let $p(Y)$ be an irreducible factor of a reduced denominator of an entry of $\mathcal T(V,Y)$. Then $p$ divides $D(Z,Y)$ for $Z$ in a Zariski-dense set. Reduce $D$ modulo $p$, as a polynomial in $Z$ over the domain $\R[Y]/(p)$: this shows $p\mid D$ in $\R[Z,Y]$. That is a pure-slot factor, which Proposition~\ref{prop:rp-intrinsic-denominator} excludes. Consequently $\mathcal T(V,Y)$ is polynomial and homogeneous linear in $Y$, giving a global linear Jacobi factor and the Clifford conclusion of Proposition~\ref{prop:gramconditional}(1), a contradiction. For $V\in\mathbb RX+U$, affineness gives $M_{Y+tV}=M_Y+tC_V(Y)$. Here $C_V(Y)$ does not depend on $t$, is unchanged by translations in $\mathbb RX+U$, and is rational with possible poles only on $Q$. Comparison of the quadratic coefficient gives $C_V\Lambda C_V^T=\Jac_V$. Choose a fixed real factor $M_V^0$ of $\Jac_V$. Then $C_V(Y)=M_V^0R_V(Y)$ with $R_V(Y)\Lambda R_V(Y)^T=\Lambda$. Both $R_V$ and $R_V^{-1}=\Lambda R_V^T\Lambda^{-1}$ have possible poles only on $Q$. Since $R_V(Y+tV)=R_V(Y)$, the factor $M'_Y=M_YR_V(Y)^{-1}$ satisfies $M'_{Y+tV}=M'_Y+tM_V^0$ and $M'_Y\Lambda M'^T_Y=\Jac_Y$. Uniqueness in Proposition~\ref{prop:rp-intrinsic-denominator} gives $\mathcal T(V,Y)=M'_Y\Lambda M_V^{0T}$. The factor $M'$ is coherent at $V$ and has possible poles only on $Q$. It must have a genuine pole, since otherwise the tensor would be Clifford by Lemma~\ref{lem:cp-pole-free}. Its denominator divides the nonzero quadratic $D(V,Y)$, so $D(V,Y)=c(V)Q(Y)$. Nonnegativity gives $c(V)>0$ for $V\ne0$. Evaluating at one $Y_0$ with $Q(Y_0)>0$ shows that $c$ is quadratic. \end{proof} \begin{proposition}\label{prop:aq-whole-aligned} A genuine whole simple quadratic denominator cannot have an auxiliary family $A\succeq0$ whose image equals $E_X$ and whose rank is eight on a nonempty real open anchor region. \end{proposition} \begin{proof} Suppose, for contradiction, that $R$ is not Clifford. Fix a generic regular unit $X$ in the region, so that $D(X,\cdot)$ is the actual reduced denominator. Image inclusion extends by polynomial continuation of the spectral projector identities. Lemma~\ref{lem:aq-common-coefficient} gives $A_z=M_zCM_z^T$ with $C>0$, and $Q(Y)=v^TCv$. The inequality $\lambda_{\min}(C)\|M_z\|_F^2\le\operatorname{tr}A_z$ bounds the partner on the regular real unit sphere. Write $F(v,u)=N(v,u)/Q(v)$ with $N$ homogeneous cubic. Under $v\mapsto tv$, its terms of bidegrees $(0,3)$ and $(1,2)$ would give poles of orders two and one. Boundedness forces both terms to vanish. Thus $F$ is affine in all seven $u$ directions. Lemma~\ref{lem:aq-affine-anchor} gives $A_V=c(V)A_X$ on $K$. On the regular locus, Cauchy--Binet and the original scaled spectrum give \[ e_8(A_z)=\det C\det(M_z^TM_z)=\det C\,|z|^{16}, \] where $e_8$ is the sum of principal eight-minors. This polynomial identity holds also at exceptional directions. On $K$ it gives $c(V)^8=|V|^{16}$; positivity gives $c(V)=|V|^2$. Sectional symmetry of $A$ now yields \[ P(Y)CP(Y)^T=(v^TCv)I. \] At $v=0$ positivity of $C$ gives $P=0$, so $P=P(v)$. For $v\ne0$, $P(v)^TP(v)=(v^TCv)C^{-1}$. At $u=0$, the Gram determinant identity and positivity give \[ |v|^{16}=\det\bigl(F(v)^TF(v)+(v^TCv)C^{-1}\bigr) \ge\frac{(v^TCv)^8}{\det C}. \] The largest eigenvalue of $C$ is at most their geometric mean, forcing $C=cI$. Equality of determinants then gives $F(v)=0$. Affineness writes $F(v,u)=L_u(v)$, with $L$ linear in $u$ and $L_u(tv)=L_u(v)$. At $xX+u+tM_Xv$, the upper frame block is $xI+L_u(v)$ and the lower block is $tP(v)$. Taking the polynomial limit of $A=cMM^T$ as $t\to0$ gives \[ (xI+L_u(v))(xI+L_u(v))^T=(x^2+|u|^2)I. \] The transpose product is the same. Adding the lower Gram produces the full radial Gram. Proposition~\ref{prop:gramconditional}(2) gives Clifford, a contradiction; coherent-pair uniqueness removes the presumed genuine pole. The limit used only $A$, not a value of the rational frame at an exceptional direction. \end{proof} \subsection{A semidefinite aligned family excludes two quadrics} \begin{proposition}\label{prop:aq-semidefinite-two} Suppose the genuine denominator at rank eight is $Q_1Q_2$, with distinct real irreducible semidefinite quadrics. There is no nonzero quadratic curvature family $A\succeq0$ with $\operatorname{Im}A_z\subset\operatorname{Im}\Jac_z$ for every real $z\ne0$. \end{proposition} \begin{proof} By Lemma~\ref{lem:aq-common-coefficient}, write $A_z=M_zCM_z^T$, $C\succeq0$, $r=\operatorname{rank}C\ge1$. For the skew rational correction $\Omega$, the mixed coefficient of $A$ implies that $\Omega HC-CH\Omega$ is polynomial. Choose a constant linear complement to the kernel of this linear map on skew matrices. The component of $\Omega$ in that complement is polynomial; subtracting it leaves the whole polar part satisfying \[ \Omega HC=CH\Omega,\qquad P\Omega\text{ polynomial}. \] Put $s=8-r$ and $k=\dim\ker(H|_{\operatorname{Im}C})$. Constant real changes of coefficient coordinates give, on spaces of dimensions $r-k,k,k,s-k$, \[ C=\operatorname{diag}(I_{r-k},I_k,0_k,0_{s-k}),\qquad H=\begin{bmatrix}H_0&0&0&0\\0&0&I&0\\0&I&0&0\\0&0&0&H_1\end{bmatrix}, \] where $H_0,H_1$ are nondegenerate. To obtain this, first normalize $C$ on its image, diagonalize the restriction of $H$, pair its radical with $k$ complementary directions, and remove the remaining cross terms by shears preserving $C$. Nondegeneracy gives $k\le\min(r,s)\le4$. Call the coefficient blocks $U_1,R,R',Z_0$, of dimensions $r-k,k,k,s-k$, and write $P_R$ for the columns of the normalized $P$ on $R$. Comparing the blocks of the commutator yields exactly \[ \Omega=\begin{bmatrix} Z&0&0&0\\0&B&M&T\\0&M&0&0\\0&-T^T&0&V \end{bmatrix},\qquad ZH_0=H_0Z, \] with $Z,B,M,V$ skew. The repeated $M$ blocks have the same sign because $M^T=-M$. The columns corresponding to the third block of $P\Omega$ are $P_RM$. By \AQisolatedref, this isolated skew block, whose support has dimension $k\le4$, has no quadratic pole. This application is unchanged by the constant, possibly nonorthogonal, coefficient normalization: if its matrix is $G_0$, embed a transformed isolated block $\Gamma$ back as $G_0\Gamma G_0^T$. This matrix is skew in the original coefficient space and its product with the original $P$ is polynomial. Restricting to an orthonormal basis of its fixed real support gives precisely the hypotheses of \AQisolatedref. Subtract the polynomial block. The remaining polar blocks are $Z$ and the skew block on $R+Z_0$, of dimension $s=8-r$. Diagonalizing $H_0$ splits $Z$ into blocks on its nonzero eigenspaces. Each is definite for $H$, hence has dimension at most seven because $H$ is indefinite; the last block also has dimension at most seven. These blocks have disjoint coefficient supports with total dimension $(r-k)+s\le8$. Each isolated product with its column subpencil of $P$ is polynomial. The same isolated-skew proposition therefore permits at most one pole-bearing block, of size six or seven. Both $Q_1,Q_2$ must occur in it. At each prime, skewness requires at least two positive Smith exponents of that generically injective linear subpencil (Lemma~\ref{lem:dq-block-degree} with $e=1$). Thus $Q_1^2Q_2^2$ divides every maximal minor of degree at most seven, a contradiction to its degree eight. No positivity over a pole function field is used. \end{proof} \subsection{Descent and alignment of the two components} We make explicit a factor-descent principle. Work over the generic anchor field $\mathbb R(X)$. A Galois-invariant collection of geometric factors of total $Y$-degree two descends, after coefficient normalization and primitive Gauss descent, to a real polynomial factor $F(X,Y)$ of $D$. Every irreducible factor of a bihomogeneous $D$ is bihomogeneous. Under $Y\mapsto Y+tX$, a factor maps to a factor of $D$, while its highest $Y$-degree part is unchanged. Comparison with that bihomogeneous factor proves invariance; the same holds in the other slot. The two shears generate $(X,Y)\mapsto(tX,t^{-1}Y)$, forcing equal degrees, and generate $(X,Y)\mapsto(Y,-X)$, giving symmetry in bidegree $(2,2)$. This also applies to a descended product, by applying it to its irreducible factors. At a generic real anchor, $F(X,\cdot)$ is a real quadratic factor of $D_X$. By Proposition~\ref{prop:rp-denominator-types}, $D_X$ is then $Q^2$ or $Q_1Q_2$, so $F(X,\cdot)$ is proportional to a semidefinite quadric. Continuity extends semidefiniteness to all fibres. Symmetry fixes one global sign: if $F(X_0,\cdot)\ge0$ and $F(X_0,Y)>0$, then $F(Y,X_0)=F(X_0,Y)>0$, so the fibre $F(Y,\cdot)$ is nonnegative. Such $Y$ are dense. Thus $F$ is the sectional form of a nonnegative curvature family. If its factors vanish on the rational kernel subspace at generic anchors, this family is aligned with $E_X$. Generic factor covers preserve algebraic invariants; no claim that real irreducibility persists at every specialization is made. \begin{proposition}\label{prop:aq-two-structure} In a surviving generic two-quadratic branch, both quadrics are aligned with $K_X$, their intrinsic global denominator is irreducible over $\mathbb R[X,Y]$, and their geometric factors have uniform intrinsic algebraic invariants. The case in which both quadrics have rank eight is excluded. \end{proposition} \begin{proof} The determinant budget is tight: $\det P=cQ_1^2Q_2^2$. At $v=0$ the radial row is zero, so $Q_1(0,u)Q_2(0,u)=0$ identically. One of the two restrictions is identically zero, and semidefiniteness puts that real subspace in its radical. At least one quadric is therefore aligned. If exactly one is aligned, distinguish the geometric factors that vanish on $K_X$. This is a rational, Galois-invariant condition. A rank-two real semidefinite quadric contributes either both of its conjugate linear factors or neither: a complex linear form vanishes on a real subspace exactly when its conjugate does. Thus the aligned collection has total degree two even in this split case. The preceding descent gives a nonzero nonnegative aligned family, contradicting Proposition~\ref{prop:aq-semidefinite-two}. Both components are aligned. If $D$ had a proper real factor, its specialization on the branch would divide $Q_1Q_2$. With pure-slot factors excluded, its degree must be two. The same shear and sign argument again produces a forbidden aligned family. Hence $D$ is irreducible over $\mathbb R[X,Y]$, and its primitive fibre is irreducible over $\mathbb R(X)[Y]$. Its geometric factors form one Galois orbit. After passage to the factor cover their degrees, ranks, support dimensions and incidence invariants are equal; these equalities persist on the generic strata. In particular a mixed geometric degree list $2,1,1$ is impossible. A geometrically irreducible quartic is also impossible: absolute irreducibility persists on a nonempty Zariski-open set of specializations, which meets the real open branch, and on that branch $D_X=Q_1Q_2$. The possibilities are two quadratics or four lines. Finally suppose both aligned quadrics have rank eight. Then $\det P(v,u)=cQ_1(v)^2Q_2(v)^2\ne0$ whenever real $v\ne0$. For a fixed nonzero real coefficient $a$, the map $v\mapsto P(v,0)a$ is an isomorphism of eight-spaces. If $P(0,u)a\ne0$, choose its unique cancelling $v\ne0$; then $P(v,u)a=0$, contradicting the determinant identity. Thus $P(0,u)=0$. The full transpose image is $E_X$, which supplies the closed-kernel condition. On the real open branch Lemma~\ref{lem:m03-closed-kernel} and Proposition~\ref{prop:gramconditional}(2) give Clifford, excluding the genuine poles. \end{proof}