\subsection{A mixed chart with a complete formal proof}\label{sec:mixedformal} One of the previously analytic exclusions admits a direct formal proof. It is the whole family in the next proposition, not a generic choice of its coefficients. \begin{proposition}\label{prop:mixedformal} For every $a\in\C$ and every $G\in\C[Z,T]$ of total degree at most three, the polynomial \[ P=W(a+T+Z^2)+G(Z,T) \] has a nonzero positive pure Gaussian moment. \end{proposition} \begin{proof} We construct the exact formal seed functional and apply Corollary~\ref{cor:leadingobstruction}. No analytic integral is needed. \emph{The formal source.} The circular fixed point is $g=t(a+T+g^2)$, with determinant $D=1-2tg$. Make the formal Gaussian change \[ T=Y-t^2(a+Y)^2,\qquad z=t(a+Y). \] It has $g=z$ and $dT/dY=D$. For any change $T(Y)=Y+O(t)$ with polynomial formal coefficients, the identity \[ \E_T[F(T)]=\E_Y\!\left[F(T(Y))T'(Y) e^{(Y^2-T(Y)^2)/2}\right] \] holds coefficientwise. To prove it, use the homotopy $T_u=Y+u(T-Y)$: its derivative is the Gaussian expectation of $(\partial_Y-Y)\{F(T_u)\dot T_u e^{(Y^2-T_u^2)/2}\}$, which is zero. Thus the determinant cancels and gives \[ \Phi_1=\E_Y[e^\Psi],\qquad \Psi=tG(z,Y-z^2)+Yz^2-z^4/2. \] Define \[ F=z/t-a-z^2,\qquad \varphi(t,z)=tG(z,F)-F^2/2, \] and let \[ M=K[z]/(\partial_z+\varphi_z)K[z],\qquad \nabla[q]=[q_t+(\varphi_t-1/t)q],\qquad K=\C(t). \] These are vector-space quotients. The formal map $J(q)=\E_Y[q(t(a+Y))e^\Psi]$ annihilates $(\partial_z+\varphi_z)K[z]$ by Gaussian integration by parts. Differentiating at fixed $Y$ differs from applying $\nabla$ by $J((\partial_z+\varphi_z)(zq/t))=0$. So $J$ is horizontal and $J(1)=\Phi_1$. The term $-1/t$ is the normalization $dz/t=dY$; removing it changes the problem. Leading-degree reduction gives $\dim_K M=\deg_z\varphi-1$. \emph{The exhaustive degree split.} Put \[ V(z)=G(z,-a-z^2),\quad L(z)=G_T(z,-a-z^2),\quad B_1=z^2G_{TT}(z,-a-z^2)/2+z(a+z^2). \] If $s=[T^3]G$, direct Taylor expansion gives \begin{equation}\label{eq:mixedphi} \varphi=tV+B_0+B_1/t+B_2/t^2, \quad B_0=zL-(a+z^2)^2/2,\quad B_2=sz^3-z^2/2. \end{equation} To check every degree drop, write $v_1=[ZT^2]G$, $v_0=[T^2]G$ and $u_2=[Z^2T]G$. Then \[ [z^6]V=-s,\qquad [z^5]V=v_1,\qquad [z^4]V=v_0-u_2-3as. \] If $n=\deg V\geq4$, the degree of $\varphi$ is $n$ and each $B_j$ has degree at most $n$. If $n\leq3$, or $V$ is constant, these three identities force $s=v_1=0$ and $u_2=v_0$. Hence $B_0=-z^4/2+O(z^3)$, $\deg B_1\leq3$ and $B_2=-z^2/2$. In that case $\deg_z\varphi=4$. There is no omitted higher-degree term tending to zero at infinity. If $n\geq4$, the leading coefficient of $\varphi$ has a nonzero $tV_n$ term and therefore has order $-1$ at infinity. If $n\leq3$ or $V$ is constant, its leading coefficient is exactly $-1/2$ and has order zero. In both cases $h=1/t$ has order one, so \[ v_\infty(h^D/a_D)>0. \] The leading-coefficient obstruction excludes $J(1)=1$. The degree calculation was exhaustive, so this proves the proposition even at multiple critical points and degree drops. The full formal frame, including every escaping dimension, is supplied by Theorem~\ref{thm:stationaryballs} rather than a new case-by-case saddle calculation. \end{proof}