\section{Positive quadratic models and the factorial case} \label{sec:positive-factorial} Throughout, $\nu=8$ and the anchor is regular, as in Section~\ref{sec:db}. Proposition~\ref{prop:db-single} supplies the following input: a real-defined, geometrically integral, full-span residue fivefold is a complete intersection $A=V(h,f)\subset\mathbb P^7_{\mathbb C}$ of two real quadrics. Here $h$ is nondegenerate and indefinite, and $A$ has no real point. We first normalize this intersection, then treat the case in which every relative eigenspace has dimension at most three. \begin{proposition}\label{prop:pf-positive-model} There is a positive definite real quadric $q$ with $\operatorname{span}_{\R}\{h,f\}=\operatorname{span}_{\R}\{h,q\}$. Every eigenvalue of $h$ relative to $q$ has multiplicity at most five. Both $A$ and its affine cone are normal. \end{proposition} \begin{proof} The joint range of $(h,f)$ on the real unit sphere is compact, and it is convex by Brickman's theorem~\cite[Theorem~1.1]{MartinezLegaz2005}. It avoids the origin, since $A$ has no real point. If $v$ is its point closest to zero, the supporting inequality $\langle v,w\rangle\ge\langle v,v\rangle>0$ for every point $w$ of the range gives a positive definite combination $q=\alpha h+\beta f$. Indefiniteness of $h$ forces $\beta\ne0$. Choose real coefficient coordinates with \[ q=\sum_{i=1}^8 a_i^2,\qquad h=\sum_{i=1}^8\lambda_i a_i^2. \] The $\lambda_i$ are real and nonzero. If an eigenvalue $\lambda$ had multiplicity at least six, the nonzero quadratic $h-\lambda q$ would have rank at most two. Over $\mathbb C$ it factors into two linear forms, possibly equal. The ideal $(h,f)=(h,f_A)$ is the prime ideal of the affine cone of $A$ (Section~\ref{sec:db}). Primality would put one linear factor in that ideal, contradicting full span. The Jacobian criterion identifies the projective singular locus as the union, over relative eigenspaces $W_\lambda$, of \[ \mathbb P(W_\lambda)\cap V(q). \] An eigenspace of dimension $m\ge2$ contributes dimension $m-2$; one of dimension one contributes nothing. Thus the projective singular locus has dimension at most three. The affine cone has dimension six, and its singular locus, including the vertex, has dimension at most four. Both spaces are regular in codimension one. As integral complete intersections they are Cohen--Macaulay, hence satisfy Serre's second condition. Serre's normality criterion applies. \end{proof} These multiplicities belong to the coefficient pencil $(h,q)$, not to the original Jacobi spectrum. Normalizing $q$ does not identify it with the original coefficient metric. In particular, a radial coefficient map remains an isomorphism and need not remain an isometry. \begin{proposition}\label{prop:pf-factorial-cone} If every relative eigenspace has dimension at most three, the standard graded cone ring \[ S=\mathbb C[a_1,\ldots,a_8]/(h,q) \] is a unique factorization domain. \end{proposition} \begin{proof} The preceding singular-locus calculation now bounds its affine dimension by two. Since $\dim S=6$, every localization at a prime of height at most three is regular, and therefore factorial \cite[Ch.~2, Theorem~1.6, p.~40]{Samuel1964}. Let $\mathfrak m=S_{>0}$ be the vertex maximal ideal. The local complete-intersection factoriality theorem applies to $S_{\mathfrak m}$: its localizations at primes of height at most three are factorial. We use the form of this theorem recorded in \cite[\S8, Remark~3 following Theorem~8.2 (Grothendieck's theorem), p.~27]{Samuel1964}. Finally, the localization map \[ \operatorname{Cl}(S)\longrightarrow \operatorname{Cl}(S_{\mathfrak m}) \] is an isomorphism for this positively graded normal domain \cite[Proposition~7.4, p.~21]{Samuel1964}. The target is zero, so the normal Noetherian domain $S$ is factorial. \end{proof} \subsection{Extraction from principal isotropic lines} We give the homogeneous extraction in a form also usable after the local class-group calculations for larger eigenspaces. All pairings on an image below are the ones induced by the symmetric matrix as an isomorphism from the dual image; they are not assumed to be restrictions of the ambient Euclidean pairing. \begin{lemma}\label{lem:pf-principal-extraction} Let $S_{\R}$ be a geometrically integral, normal, standard graded real domain with degree-zero part $\R$, and put $S=S_{\R}\otimes_{\R}\mathbb C$. Let $C$ be a symmetric matrix of homogeneous degree two over $S_{\R}$ with generic rank two and hyperbolic induced pairing over $\operatorname{Frac}(S_{\R})$. Suppose the saturated rank-one modules of its two generic complex isotropic lines in $S^N$ are principal. Then, over $S_{\R}$, $C$ has one of the forms \begin{equation}\label{eq:pf-extracted} C=uu^T-vv^T, \qquad\text{or}\qquad C=c b^T+b c^T, \end{equation} where $u,v$ are homogeneous linear columns, or $c$ is a constant column and $b$ is homogeneous quadratic, respectively. \end{lemma} \begin{proof} The scaling $\sigma_t:a\mapsto ta$, $t\in\mathbb C^*$, multiplies $C$ by $t^2$ and so permutes the two isotropic lines. The stabilizer of a line has index at most two, and the divisible group $\mathbb C^*$ has no subgroup of index two. Each saturated module is therefore $\sigma_t$-stable for all $t$, and comparison of homogeneous components shows that it is graded. A principal graded rank-one module has a homogeneous generator; for example, choose a homogeneous generator of its one-dimensional fiber at $S_{>0}$ and use graded Nakayama. Denote the two polynomial vector generators by $p,r$, of nonnegative degrees $d_p,d_r$. Over the fraction field, symmetry and isotropy give \[ C=t(pr^T+rp^T),\qquad 0\ne t\in\operatorname{Frac}(S). \] At each height-one prime, saturation makes both vectors primitive. Their reductions are nonzero, and their symmetric cross product is nonzero over the residue field: this holds both for independent vectors and for proportional vectors in characteristic zero. Some entry of $pr^T+rp^T$ is therefore a unit locally. Since $C$ is polynomial, $t$ has no negative height-one valuation. Normality gives $t\in S$. Homogeneity then yields \begin{equation}\label{eq:pf-degree-budget} \deg t+d_p+d_r=2. \end{equation} This step includes any common divisorial scalar factor of $C$. Hyperbolicity over the real fraction field makes the two lines individually real-defined. Conjugation preserves each saturated module, so it changes its homogeneous generator by a unit of $S$. Every such unit is a nonzero complex constant. Rescaling the generator makes it real: conjugating twice gives modulus one for that constant, which is removed by a scalar phase. Thus $p,r$ and then $t$ can be taken real. If $d_p,d_r>0$, both are one and $t$ is constant. Absorb $t$ into one column, then use their real sum and difference to obtain the first form in \eqref{eq:pf-extracted}. Otherwise one generator is constant, and its partner multiplied by $t$ is quadratic by \eqref{eq:pf-degree-budget}, giving the second form. \end{proof} \begin{proposition}\label{prop:pf-factorial-normal-form} In the factorial case of Proposition~\ref{prop:pf-factorial-cone}, suppose $C=B(a)B(a)^T$ is the actual quadratic transpose Gram (the matrix $G$ of Lemma~\ref{lem:si-incidence}), of generic rank two on $A$, whose generic transpose kernel is a totally isotropic three-space for the standard eight-form. Then, as an ambient real quadratic matrix, it has one of the forms \begin{align} C&=hD_0+qD_1+uu^T-vv^T,\label{eq:pf-balanced}\\ C&=hD_0+qD_1+c b^T+b c^T,\label{eq:pf-unbalanced} \end{align} where $D_0,D_1$ are constant real symmetric matrices and the columns have the degrees specified in Lemma~\ref{lem:pf-principal-extraction}, with $c\ne0$ in \eqref{eq:pf-unbalanced}. \end{proposition} \begin{proof} Let $K_0=\ker B$ be the generic transpose kernel over the real function field of $A$. Since $\operatorname{Im}B^T=K_0^\perp$, we have $\operatorname{Im}C=B(K_0^\perp)$, and for $y=Cx$, $y'=Cx'$ the induced pairing is $x^TCx'=(B^Tx)^T(B^Tx')$. Thus $B$ induces an isometry from $K_0^\perp/K_0$, with the standard form, onto $\operatorname{Im}C$ with its induced pairing. Splitting off the three hyperbolic planes associated with the isotropic three-space leaves determinant square class $-1$ from the standard eight-form. A nonsingular binary form of this determinant class is hyperbolic. The same holds on the affine cone, whose function field is a purely transcendental extension by the scaling parameter. The saturated modules of the two isotropic lines are rank-one reflexive modules over the normal domain $S$: equivalently, they are the intersections of their height-one localizations inside their fraction-field lines. Factoriality makes their divisor classes zero, so they are principal. Since $S$ is normal, so is its conjugation-fixed subring $S_\R$, and Lemma~\ref{lem:pf-principal-extraction} applies. Lift its homogeneous columns to ambient real polynomials. The difference between $C$ and the lifted expression has degree two and belongs to the ideal $(h,q)$. It is therefore $hD_0+qD_1$ with constant real symmetric coefficient matrices, as asserted. \end{proof} For the actual full eight-coefficient pencil, Proposition~\ref{prop:zs-unbalanced} applies to \eqref{eq:pf-unbalanced}. Proposition~\ref{prop:balanced-alignment} applies to \eqref{eq:pf-balanced}: relative multiplicities at most three imply $\lambda_2<\lambda_7$. For the unbalanced form, the coefficient space $A$ of Proposition~\ref{prop:zs-unbalanced} is the eight-dimensional coefficient space, as in Sections~\ref{sec:balanced-alignment} and~\ref{sec:zs-tools}. Thus the full image equals the Jacobi image at every anchor of this case. On a real-open branch of such anchors, the frame-free closed-kernel criterion (Lemma~\ref{lem:m03-closed-kernel}) and Proposition~\ref{prop:gramconditional}(2) make $R$ Clifford; see Section~\ref{sec:assembly}. No smoothness or factoriality assertion for relative multiplicities four or five is used here; their local class-group detectors are required in the subsequent section.