\subsection{Quadratic elimination and the repeated lines} \label{sec:repeatedlines} The final two leading types have a repeated factor. They reduce to a polynomial in one variable after eliminating two Gaussian variables. The full leading-coefficient theorem makes an escape table unnecessary: only the actual degree, its leading coefficient and the determinant factor are needed. \begin{lemma}[The determinant factor in quadratic elimination] \label{lem:quadraticelimination} Over $\C(t)$, let a phase have the form \[ -\tfrac12 y^T M(t)y+L(t,x)^Ty+V(t,x), \] where $M$ is an invertible symmetric matrix and $L,V$ are polynomial in $x$. Eliminating $y$ identifies its top twisted module and constant volume seed with the one-variable phase and normalization \[ \varphi=V+\tfrac12L^TM^{-1}L,\qquad h=(\det M)^{-1/2}. \] The parameter connection is $\partial_t+\varphi_t+h'/h$. An algebraic square root is allowed. If the actual degree $d$ of $\varphi$ is at least two and $d\,v_\infty(h)>v_\infty([x^d]\varphi)$, its constant seed cannot have a rational unit quotient. \end{lemma} \begin{proof} The change $y=z+M^{-1}L$ has Jacobian one. Its spatial twisted relations reduce the $z$ factor to the one-dimensional Gaussian quotient, with normalized contractions $\ell(z_i z_j)=(M^{-1})_{ij}$. Parameter differentiation of its constant class inserts $-z^TM'z/2$, hence \[ -\tfrac12\operatorname{tr}(M^{-1}M')=h'/h. \] Terms from the moving center are spatial exact forms, by the product rule or Cartan's formula. This proves the identification of both connection and seed. After a finite cover at infinity, $h$ is a Laurent series and Theorem~\ref{thm:stationaryballs} applies. The cover multiplies both sides of the valuation inequality by the same positive integer. The elimination is over the generic field. It need not define a conditional formal average at $t=0$. The original Wick functional first gives its rational seed quotient by Lemma~\ref{lem:seedfunctional}; this quotient is transported through the displayed module identification. No image-kernel calculation or convergent original period is needed. \end{proof} \begin{lemma}\label{lem:quadraticfibrehighdegree} Let $u,Y_1,\ldots,Y_s$ be independent standard Gaussian variables and \[ P=Y^TAY+B(u)^TY+g(u),\qquad A=A^T, \] with $B,g$ arbitrary polynomials. Put \[ D(t)=\det(I-2tA),\quad C(t)=(I-2tA)^{-1},\quad \varphi=tg-u^2/2+(t^2/2)B^TCB. \] If $\deg_u\varphi\ge3$, then $P$ is not pure. \end{lemma} \begin{proof} Quadratic elimination gives $h=D^{-1/2}$. Let $d=\deg D$ and $r=\deg_u\varphi\ge3$. Its leading coefficient is $N(t)/D(t)$, where $N$ is a nonzero polynomial divisible by $t$: the degree-$r$ coefficient vanishes at $t=0$. Consequently \[ r v_\infty(h)-v_\infty(N/D) =(r/2-1)d+\deg N>0. \] Apply Lemma~\ref{lem:quadraticelimination}. This estimate is independent of the degrees of $B,g$ and the dimension $s$. \end{proof} \subsubsection*{A nonisotropic repeated factor} Normalize the repeated factor to $u$ with $\E[u^2]=1$ and take an orthogonal Gaussian complement $Y\in\C^2$. Every such cubic has the form in Lemma~\ref{lem:quadraticfibrehighdegree}, with \[ B(u)=p u^2+q u+r,\qquad\deg g\le3. \] For a double line $p\ne0$. If its effective quartic coefficient is nonzero, the lemma excludes purity immediately. If that coefficient vanishes identically, then \[ 0=p^T\operatorname{adj}(I-2tA)p =p^Tp-2t\bigl(\operatorname{tr}(A)p^Tp-p^TAp\bigr). \] Thus $p^Tp=p^TAp=0$. The nonzero vector $(0,p)$ is isotropic, lies in the leading singular plane $u=0$, and satisfies $F_2(0,p)=0$. It supplies the frame of Lemma~\ref{lem:frame}. For a triple line, $p=0$ and $[u^3]g\ne0$. The effective phase has exact cubic coefficient $t[u^3]g$, so the same high-degree lemma excludes purity. This exhausts all matrices $A$, including nilpotent and defective ones, without a matrix case split. \subsubsection*{An isotropic repeated factor} Use Gaussian circular coordinates with repeated factor $Z$. The singular plane is $Z=0$, whose only isotropic direction is the $W$ axis. Write $\lambda=[W^2]F_2$. If $\lambda=0$, the frame test applies. Assume $\lambda\ne0$. For a double line write its other factor as $aW+bT+cZ$. The Gaussian null rotation \[ Z=Z',\quad T=T'+sZ',\quad W=W'-sT'-s^2Z'/2 \] removes the $T$ term if $a\ne0$, or removes the $Z$ term if $a=0,b\ne0$. Scaling $P$ gives exactly these leading forms: \[ \mathrm I: Z^2W+a_0Z^3,\qquad \mathrm {II}:Z^2T,\qquad \mathrm {III}:Z^3. \] The last is the triple line. The rotation preserves the Gaussian and keeps the $W^2$ coefficient nonzero. All lower terms remain arbitrary. Write $P=\lambda W^2+W H(Z,T)+G(Z,T)$. In type I, $H=Z^2+bZ+cT+d$ and $G=a_0Z^3+G_2$. In types II and III, $H=bZ+cT+d$ and $G=Z^2T+G_2$ or $Z^3+G_2$ respectively. Here $G_2$ is an arbitrary polynomial of degree at most two. Eliminating $W$ in $tP-ZW-T^2/2$ gives \[ \Psi=t\left(G-\frac{H^2}{4\lambda}\right) +\frac{ZH}{2\lambda}-\frac{T^2}{2} -\frac{Z^2}{4\lambda t}. \] Its $T^2$ coefficient is $C_T=a t-1/2$, where $a=g_{02}-c^2/(4\lambda)$ and $g_{02}=[T^2]G_2$. Eliminate $T$ as well. If $e=\deg_t C_T\in\{0,1\}$, the total normalization has $v_\infty(h)=(1+e)/2$. The actual remaining leading coefficients are \[ \begin{array}{c|c|c} \text{type}&[Z^d]\varphi&d\\\hline \mathrm I&\displaystyle -\frac{t}{4\lambda}-\frac{c^2t^2}{16\lambda^2C_T} =-\frac{t(2g_{02}t-1)}{8\lambda C_T}&4\\[5pt] \mathrm {II}&\displaystyle-\frac{t^2}{4C_T}&4\\[4pt] \mathrm {III}&t&3. \end{array} \] None is identically zero. Their valuations are at most $e-1$, $e-2$, and $-1$, respectively. Each is strictly less than $d(1+e)/2$. Lemma~\ref{lem:quadraticelimination} excludes purity in all three types. In particular this includes the boundary where the quartic coefficient tends to a nonzero constant while lower coefficients grow. \begin{proposition}\label{prop:repeatedcomplete} Every pure ternary cubic with repeated leading factor has a Gaussian affine-linear coordinate. It therefore satisfies $m>3\deg Q$. \end{proposition} \begin{proof} The nonisotropic and isotropic alternatives above exhaust the repeated factor. In every surviving branch the explicit frame test succeeds. Apply the complete affine-linear theorem. \end{proof}