\subsection{Ordinary nodes with a finite affine critical set} \label{sec:ordinaryfinite} The three ordinary-node families admit one argument whenever the critical set of $P$ is finite. In particular we need not solve for each escaping critical point, assume it is Morse, or separate its action from the actions of other escaping points. The Gaussian quadratic itself detects the unwanted contribution at infinity. Retain the notation and the full pair $(P,q)$ of Lemma~\ref{lem:ordinarynormal}. Write $q_2$ for the homogeneous quadratic part of $q$. The absence of a Gaussian frame is exactly the following condition at the leading nodes: \begin{equation}\label{eq:nodenoframe} F_2(w)=0\quad\Longrightarrow\quad q_2(w)\ne0. \end{equation} Indeed the Gaussian bilinear form in these affine coordinates is $2q_2$; the value of $F_2$ at a leading singular direction is unchanged by translation. In the common normal form the $T$ node already satisfies this condition because $m\ne0$. The only extra conditions are $c\ne0$ or $B\ne0$ when $d=0$, and $b\ne0$ or $A\ne0$ when $a=0$. \begin{lemma}[The stationary action identity]\label{lem:actionidentity} Put $r=1/t$, and let $x(r)$ be an algebraic critical branch of $S=P/r-q$. A prime means total differentiation along the branch. Then \begin{equation}\label{eq:actionidentity} S'=-P(x(r))/r^2,\qquad q(x(r))=-S-rS'. \end{equation} If $x(r)$ is bounded, the polar part of $S$ is $\alpha/r$, where $\alpha=P(x(0))$. If it escapes and \eqref{eq:nodenoframe} holds, its polar part is neither zero nor $\alpha/r$ for any constant $\alpha$. \end{lemma} \begin{proof} Use a finite Puiseux cover so every branch is meromorphic. At a critical point the chain-rule term $\nabla_x S\cdot x'$ vanishes, giving both identities. For a monomial $c r^{-\kappa}$ of the polar part of $S$, its contribution to $-S-rS'$ is $(\kappa-1)c r^{-\kappa}$. Thus bounded $q(x(r))$ permits only the pole $\alpha/r$ in $S$. Its coefficient is $P(x(0))$ when $x$ is bounded. For an escaping branch let $R=\|x\|$. The critical equation is $\nabla P=r\nabla q$, so $\|\nabla P\|/R\longrightarrow0$. The leading projective direction is a singular point of $F_3$. The ordinary-node argument in the proof of the preceding coercive lemma works at this one direction: write $x=sw+y$, $y=o(s)$, in a fixed complementary plane. The injectivity there of $H=\operatorname{Hess}F_3(w)$ forces $y$ bounded from \[ \nabla P/s=Hy+\nabla F_2(w)+O((1+\|y\|)^2/|s|)=o(1). \] Pairing with $w$ then gives $F_2(w)=0$. Condition \eqref{eq:nodenoframe} gives $q_2(w)\ne0$, so $q(x(r))$ has a pole: its leading quadratic term cannot be cancelled by its lower-degree terms. Identity~\eqref{eq:actionidentity} now rules out both claimed polar parts. This argument includes all escape scales and all coincident actions. \end{proof} We specify the classical rank input before using it. This is a convergent algebraic application of formal stationary phase, not an extension to arbitrary nonconvergent coefficients or rings. \begin{lemma}[Rank accounting for a tame algebraic pencil] \label{lem:tamepencilrank} For the present no-frame pencil, shrink a punctured $r$ disk to avoid its finitely many exceptional parameters. Its top twisted de Rham module has rank \[ N=\sum_{\text{critical points of }S_r}\mu. \] After a finite cover, the rank of its block with nonzero exponential polar part $\gamma$ is the sum of these generic local Milnor numbers over critical branches with action polar part $\gamma$. Equal actions add their multiplicities. Consequently the rank of the zero-exponential block is the sum of the generic Milnor numbers of the branches with bounded action. \end{lemma} \begin{proof} For fixed small nonzero $r$, the homogeneous quadratic term of $S_r$ at a leading node $w$ is $F_2(w)/r-q_2(w)$. It is nonzero at every node, by \eqref{eq:nodenoframe} and exclusion of finitely many parameters. The ordinary-node coercivity lemma gives $\|\nabla S_r\|\ge c_r\|x\|$ outside a compact set. The critical locus is therefore finite: a positive-dimensional affine algebraic component would be unbounded. Pham's tame Milnor decomposition~\cite[Part~I, Sections~1--4]{Pham} gives the displayed rank, retaining the Milnor number of every degenerate point. Here are the checks needed to apply the direct-image rank formula. Take $U=\C_r^*\times\C_x^3$, $f=r$, $g=S_r$, and the regular algebraic module $\mathcal O_U$. Form the ordinary direct image by $(f,g)$, followed by exponential integration in the $g$ coordinate. All these are algebraic maps; $g$ is regular on $U$. The ordinary direct image is regular holonomic. Over the punctured $r$ disk its relevant cohomology is the middle, reduced Gauss--Manin module of the fibers of $S_r$. The other, constant fiber cohomology has zero exponential integral in the affine $g$ line. Punctual parameter terms disappear on the punctured disk. There are no extra finite-$g$ singular curves from spatial infinity. The coercive estimate is locally uniform on a compact subset of the punctured $r$ disk. For $g$ in a compact set, the identity $P=r(g+q)$ on a fiber gives \[ \partial_r S_r=-(g+q)/r=O(1+\|x\|^2). \] Lifting $r$ variation by the normalized gradient therefore uses a vector field of at most linear growth. Lifting $g$ variation uses a field $O(1/\|x\|)$. These fields are complete outside a fixed ball on each such parameter neighborhood. They trivialize the outer fiber pairs, so all nonvertical finite-$g$ discriminant curves come from affine critical values. Along such a curve the conormal multiplicity of the middle Gauss--Manin module is the sum of the local Milnor numbers: a transverse target disk computes precisely the corresponding isolated vanishing cycles. This also applies to persistent degenerate points and equal critical values. First make a common finite Puiseux base change of all the critical-value curves. Apply the nonramified case of Roucairol's Theorem~0.1(2), proved in her Section~1.2, and Theorem~0.3 \cite{RoucairolFormal} to this identified ordinary direct image. They give exactly the stated nonzero polar parts and ranks. Count every graph after this base change, including distinct conjugates with the same polar part. For example the two graphs $g=r^{-1}\pm r^{1/2}$ must contribute twice after $r=\rho^2$; retaining just their one polar expression would lose rank. Their rank statement requires no separation of the constant terms of equal polar parts. The additional separation hypothesis in her Theorem~0.2 concerns its monodromy formula, which we do not use. Vertical components at $r=0$ and the boundary $g=\infty$ do not add polar branches in Theorem~0.1. Subtracting the nonzero-block ranks from $N$ proves the last assertion. In particular we have not inferred the regular rank from an Euler characteristic alone. \end{proof} \begin{proposition}\label{prop:ordinaryfinite} Suppose the common ordinary-node pair satisfies \eqref{eq:nodenoframe} and $P$ has finite affine critical set. Then $P$ is not pure. More generally every original Gaussian marked period is either zero or is not a finite sum of polynomials in $t$ times exponentials $e^{\beta t}$. \end{proposition} \begin{proof} Every bounded critical branch of $S_r$ limits to a critical point of $P$. Disjoint small Milnor neighborhoods of these finitely many points contain, for the deformation $P-rq$, total critical multiplicity $\mu_p$ in the neighborhood of $p$. This is conservation of the local intersection multiplicity of the gradient, and does not require a Morse perturbation. By Lemma~\ref{lem:actionidentity} its action has polar part $P(p)/r$. Every escaping branch has nonzero polar part. Hence Lemma~\ref{lem:tamepencilrank} identifies the entire zero-exponential rank as \begin{equation}\label{eq:ordinaryzerorank} \sum_{p:\,dP(p)=0,\,P(p)=0}\mu_p. \end{equation} We construct that many independent local formal functionals and compute their monodromy over the unramified field $\C((r))$. At an isolated critical point at least one of $u,v$ is nonzero. Indeed if both vanish, criticality requires $h=e=f=0$, in which case the whole $T$ axis is critical. In the chart $u\ne0$ put \[ s=uv+h,\quad R(u)=G(u,-h/u),\quad T'=T+\frac{G(u,(s-h)/u)-G(u,-h/u)}s, \] where $G=P-T(uv+h)$. The quotient is analytic, and exactly $P=sT'+R(u)$. The coordinate Jacobian is $u\ne0$. At a critical point $p$ the one-variable critical multiplicity of $R$ is $\mu_p$. An analytic one-variable coordinate $y$, with $w=T'$, therefore gives the exact identity \[ P=\alpha+sw+y^k,\qquad k=\mu_p+1,\quad\alpha=P(p). \] The volume Jacobian and $e^{-q}$ multiply every amplitude by an analytic unit $V(y,s,w)$; no Gaussian mean or cross term has been discarded. For $j=1,\ldots,k-1$ define a formal local functional on an analytic monomial $y^a s^b w^c$ as zero unless $b=c$ and $a=j-1+\ell k$ for an integer $\ell\ge0$. Otherwise its value is \begin{equation}\label{eq:ordinarycharacters} e^{\alpha/r}r^{1+j/k} (-1)^{b+\ell}b!\,(j/k)_\ell\,r^{b+\ell}. \end{equation} Evaluate a polynomial form after multiplying its transformed amplitude by $V$. At each power of $r$ only finitely many Taylor terms contribute. The recurrences \[ I_{a+k}=-\frac{a+1}{k}r I_a,\qquad J_{b+1,b+1}=-(b+1)rJ_{b,b} \] prove spatial integration by parts and parameter horizontality. Thus these are functionals on the full twisted quotient, not on an assumed cyclic image. They exist even if the corresponding asymptotic series diverge; their input coefficients here are the Taylor series of the stated analytic coordinates. They are independent. Prescribe, at each $p$, the jets $1,y,\ldots,y^{k-2}$ on the line $s=w=0$ and zero jets at the other points, using polynomial representatives of finite local jets and the Chinese remainder theorem. After dividing column $j$ by $e^{\alpha/r}r^{1+j/k}$, the leading evaluation matrix is the jet map multiplied by the analytic unit $V$. It is triangular with nonzero diagonal. Its higher terms are in $r\C[[r]]$. For $\alpha=0$ these independent characters fill exactly the rank in \eqref{eq:ordinaryzerorank}; there is no unaccounted regular piece. Their formal monodromy eigenvalues, or their inverses if one uses the dual module, are \[ \exp(2\pi i j/k),\qquad 1\le j