\section{The first tensor shell and doubling}\label{sec:shell} The first gauge determines the slopes. Exact formulas also require a description of its minimizers and a gap to other lattice points. \begin{theorem}[Signed rectangles and a gap]\label{thm:shell} Let $p_i\ge2$ and \[ R=\bigotimes_{i=1}^w[I_{p_i-1}\mid-\one],\qquad Z=R[0,1]^{\prod_i p_i},\quad \mu=2^{1-w},\quad \nu=\frac{2}{2^w-1}. \] The nonzero integer points of gauge less than $\nu$ are precisely \[ b=\varepsilon\one_{S_1\times\cdots\times S_w}, \qquad\varepsilon\in\{1,-1\},\quad \varnothing\ne S_i\subseteq\{0,\ldots,p_i-2\}. \] All have gauge $\mu$. Every other nonzero integer point has gauge at least $\nu$, without a claim that $\nu$ is the actual second value. \end{theorem} \begin{proof} Let $\rho_a$ be the rows of $R$. For a zero-sum vector $v$, set $h(v)=\sum_r\max(v_r,0)$, so that $v\cdot x\le\lambda h(v)$ whenever $0\le x_r\le\lambda$. If $a,b$ differ in $e$ coordinates, their rows share $2^{w-e}$ entries with the same sign. Hence \begin{equation}\label{eq:row-difference} h(\rho_a-\rho_b)=2^w-2^{w-e}\le2^w-1\quad(a\ne b). \end{equation} If $c$ is obtained by replacing one coordinate of $a$ by that of $b$, then also \begin{equation}\label{eq:row-recombination} h(\rho_a+\rho_b-\rho_c)\le2^w-1. \end{equation} When that coordinate changes, write $\rho_a=U\otimes(e_j-e_*)$, $\rho_b=V\otimes(e_k-e_*)$, and $\rho_c=U\otimes(e_k-e_*)$. The three sections are $U,V-U,-V$; their total $\ell^1$ norm is $2^{w+1}-2^{w-e+1}$. If the coordinate is unchanged, the combination is simply $\rho_b$. Now let $0<\|b\|_Z<\nu$. The single-row bound gives $|b_a|<2$, so every integral coordinate is $-1,0,1$. Opposite signs would contradict \eqref{eq:row-difference}, since $2\le(2^w-1)\|b\|_Z<2$. After reversing sign, all nonzero values are 1. If $b_a=b_b=1$ but a recombination $c$ has value zero, \eqref{eq:row-recombination} gives the same contradiction. Thus the support is closed under one-coordinate recombination. Replacing coordinates one at a time shows that it is the product of its nonempty projections. Conversely, let $f_i$ equal 1 on $S_i$ and $-1$ elsewhere, including the anchored index $p_i-1$. Then $[I\mid-\one]f_i=2\one_{S_i}$, and \[ x=\frac{\one+\varepsilon\bigotimes_i f_i}{2^w} \] has coordinates 0 or $\mu$ and satisfies $Rx=b$. The lower bound $\mu$ follows from a nonzero integral coordinate and the row bound $2^{w-1}\lambda$. \end{proof} For $t=\prod p_i^{a_i}$, the CRT matrix is, up to permutations, a direct sum of $h=\prod p_i^{a_i-1}$ copies of this model. The body is a product, and its gauge is the maximum of the block gauges. On the \emph{integral} branch, each block of a minimum point is zero or a signed rectangle, with at least one nonzero block. Their number is \[ \left(1+2\prod_i(2^{p_i-1}-1)\right)^h-1. \] This statement does not silently include ramified cosets. \begin{proposition}[Doubling an odd order]\label{prop:doubling} For odd $m>1$, choose $\xi=-\zeta_m$ of order $2m$. The two rings coincide, and in the same integral basis \[ R_{2m}[:,r]=(-1)^rR_m[:,r\bmod m],\qquad Z_{2m}=2Z_m. \] The two matrices have the same absolute values of nonzero minors. If $T$ represents multiplication by $\zeta_m+\zeta_m^{-1}$, the order-$2m$ operator is $(B^2+1)I+BT$. \end{proposition} \begin{proof} The reordered columns are $[R_m,-R_m]$. Since $Z_m$ is centrally symmetric, their zonotope is $Z_m+(-Z_m)=2Z_m$. A minor using two copies of one column is zero; the other minors differ from those of $R_m$ only in sign and order. Every minor of $R_m$ occurs. Finally, $(B-\xi)(B-\xi^{-1})=(B+\zeta_m)(B+\zeta_m^{-1})$. \end{proof} Capacity transport is a separate matter. For $N_r=\max\{0,1+\lfloor(k-1-r)/m\rfloor\}$, the coordinate \[ z_r=\sum_{q\ge0,\ r+qm