\subsection{A polynomial phase with logarithmic volume}\label{sec:puncturedpoly} The puncture itself can be retained in the formal construction when the phase is polynomial. This gives a stronger bound than the one for $dx$: logarithmic volume does not change under a scaling of $x$. \begin{theorem}\label{thm:puncturedpoly} Let $K=\C((r))$, $\varphi\in K[x]$, and $h\in K\setminus\{0\}$. On \[ H=K[x,x^{-1}]/(x\partial_x+x\varphi_x)K[x,x^{-1}],\qquad \nabla[q]=[q_r+(\varphi_r+h'/h)q], \] there is, after finite ramification, a full frame of rank-one formal exponential functionals. Every zero-exponential functional in this frame has seed value of valuation at least $v(h)$. Thus if $v(h)>0$, the differential seed submodule has no horizontal quotient taking $[1]$ to one. This includes a constant phase. \end{theorem} \begin{proof} Write $D=\deg\varphi$, taking $D=0$ for a constant phase, including zero. If $D\geq1$, the quotient has basis $1,x,\ldots,x^{D-1}$. Indeed, on a negative monomial the lowest term of $x\partial_x+x\varphi_x$ is its nonzero integer exponent times that monomial. This eliminates all negative powers without introducing lower ones. On polynomials, successive leading terms eliminate all degrees at least $D$. Conversely an exact polynomial cannot come from a Laurent polynomial with a negative lowest term, and a nonzero polynomial preimage of degree $m$ has exact degree $m+D$. This proves independence. For $D=0$, the quotient of $x\partial_x$ has the single basis element $1$. Use the critical balls constructed in Theorem~\ref{thm:stationaryballs}. At most one contains the origin. Every other ball has $\rho_B>v(\xi_B)$, so $\gamma_B=\delta_B/\xi_B$ has positive valuation. With the same $c_B,F_B,E_B$ and fixed functionals as before, the local logarithmic functionals are \begin{equation}\label{eq:puncturedaway} q\longmapsto h e^{c_B}\frac{\delta_B}{\xi_B} \ell^0_{B,j}\!\left( \frac{e^{E_B}q(\xi_B+\delta_B y)}{1+\gamma_B y}\right). \end{equation} The geometric denominator and all negative powers in $q$ expand coefficientwise because $v(\gamma_B)>0$. Every coefficient is a polynomial in $y$. The normalized logarithmic Jacobian is $dx/x=\delta_B\,dy/(\xi_B+\delta_B y)$. We retain the puncture in a separate central ball. If a critical ball contains zero, write it as $v(x)\geq\rho_0$ and put $\delta_0=r^{\rho_0}$. Translation within that ball gives \[ \varphi(\delta_0 y)=c_0+F_0(y)+E_0(r,y),\qquad c_0=\varphi(0),\quad E_0\in r\C[[r]][y],\quad F_0(0)=0. \] Here $d_0=\deg F_0\geq2$, and the ball contains $d_0-1$ critical roots with multiplicity. If no critical ball contains zero, choose $\rho_0$ sufficiently large that the central ball is disjoint from all critical balls and every nonconstant coefficient of $\varphi(\delta_0 y)$ has positive valuation. Then $F_0=0$. Such a choice exists: there are finitely many roots and coefficients. This also handles $D\leq1$ without any critical balls. For $d_0\geq2$, the fixed punctured quotient \[ H_0=\C[y,y^{-1}]/(y\partial_y+yF_0')\C[y,y^{-1}] \] has basis $1,y,\ldots,y^{d_0-1}$, by the same reduction argument. For $F_0=0$, it has basis $1$ and dual functional the constant term. For each element $\lambda_j$ of the dual basis define \begin{equation}\label{eq:puncturedcentral} q\longmapsto h e^{c_0}\lambda_j \!\left(e^{E_0}q(\delta_0 y)\right). \end{equation} There is no factor $\delta_0$: $dx/x=dy/y$. Each coefficient is a Laurent polynomial, so the expression is well defined. We check both identities, including parameter dependence. Spatial exactness in the central ball is conjugation by $e^{E_0}$ of $y\partial_y+yF_0'$. The derivative discrepancy is $\delta_0'/\delta_0$ times the same exact expression applied to $e^{E_0}q$. On an outside ball put $x=\xi_B+\delta_B y$ and $w=\xi_B'+\delta_B'y$. After removing $h e^{c_B}$, the derivative discrepancy in \eqref{eq:puncturedaway} is \[ e^{E_B}\left[ \left(\frac{\delta_B'}x-\frac{\delta_B w}{x^2}\right)q +\frac{\delta_B w}x(q_x+\varphi_xq)\right] = (\partial_y+F_B')\left(e^{E_B}\frac wxq\right). \] It is annihilated by $\ell^0_{B,j}$. The identical product rule proves spatial exactness there. Thus every displayed functional is horizontal for the stated connection. Completeness follows from polynomial separation of these disjoint balls, not merely their number. For the central ball use the monomials of the basis of $H_0$; for each outside ball use $1,\ldots,y^{d_B-2}$. Lemma~\ref{lem:ballseparation} gives polynomial amplitudes with exactly these reductions on their own ball and zero on the others. Normalize the central columns by $h e^{c_0}$ and the outside columns by $h e^{c_B}\delta_B/\xi_B$. Their matrix reduces to the identity. If the central ball contains critical roots, it contributes their multiplicity plus one; otherwise it contributes one. Together with all outside roots this gives $D$ dimensions for $D\geq1$, or one for $D=0$, the full quotient rank. As in Theorem~\ref{thm:stationaryballs}, this frame is a direct sum of rank-one exponential factors after ramification. For a zero factor, \eqref{eq:puncturedcentral} has seed valuation at least $v(h)$; \eqref{eq:puncturedaway} has valuation at least $v(h)+\rho_B-v(\xi_B)>v(h)$. Semisimplicity now extends a quotient of the differential seed submodule to the full module. Nonzero exponential factors cannot map horizontally to the Laurent field. A constant combination of the zero-factor seed values cannot be one if $v(h)>0$. \end{proof} \begin{remark}[Why two irregular ends require another argument] For $\varphi=r(x+x^{-1})$ and $h=1$, put $\theta=r\partial_r$. Every logarithmic period satisfies \[ (\theta^2-4r^2)I=0. \] Indeed $y=x+x^{-1}$ and $z=x-x^{-1}$ obey $(x\partial_x+x\varphi_x)z=y+rz^2$ and $y^2-z^2=4$. These identities give the equation by integration by parts. The quotient has rank two, and $[1],[y]$ are independent: its basis is $x^{-1},1$ and $[y]=2[x^{-1}]$. An explicit formal basis of scalar solutions is \[ I=\sum_{n\geq0}\frac{r^{2n}}{(n!)^2},\qquad J=I\log r-\sum_{n\geq1}\frac{H_n r^{2n}}{(n!)^2}, \quad H_n=\sum_{j=1}^n\frac1j. \] The coefficient recurrence $n^2a_n=a_{n-1}$ proves the first equation; $H_n-H_{n-1}=1/n$ proves the second. Their $\theta$-Wronskian is one: its derivative is zero by the equation and its constant term is one. Thus the logarithm is a full rank-two phenomenon, with nontrivial unipotent monodromy that no finite ramification removes. These are the familiar modified Bessel solutions; see~\cite[10.25.1, 10.31.2]{DLMF}. This example does not refute the desired logarithmic bound: multiplication by $h=r$ makes both seed values tend to zero. It does refute an extension of the semisimple polynomial-frame theorem to all Laurent phases. The remaining proof must allow such extensions. \end{remark} The residue dimension at the puncture is essential. Dropping it would give rank $D-1$ instead of $D$ and an incomplete frame. This theorem still permits nonconvergent formal coefficients, but is a valued-field statement. It does not cover a phase with both positive and negative powers of $x$. \begin{corollary}\label{cor:oneend} Suppose $\varphi(t,x)$ is rational in $t$ and polynomial either in $x$ or in $x^{-1}$. For the logarithmic-volume connection normalized by $h=1/t$, no horizontal formal functional at another point can have seed value one. \end{corollary} \begin{proof} For a polynomial in $x$, apply Theorem~\ref{thm:puncturedpoly} at $r=1/t=0$, where $v(h)=1$. The rational seed restriction is justified as in Corollary~\ref{cor:leadingobstruction}. For a polynomial in $x^{-1}$ use $y=1/x$; $dx/x=-dy/y$ only changes a constant factor. \end{proof}