\subsection{Projection after the slow decay} \label{sec:small-section-delayed-projection} The physical phase at the end of the longer interval must now be expressed in the coordinates of the growing core. The projection divides by a height of order $a^2$. Its radial response is controlled because the strong wall and the decaying weak wall have both reached order $a^2$, and their ordinary derivatives have reached order $a$. These are precisely the scales needed to reuse the compact comparison of Section~\ref{sec:small-section-projection}. We keep the positive-wall family \eqref{eq:slow-variation-label-domain}, with $q>2$, $D>2$, and $P>6D+2$. Write its endpoint as \begin{equation} z_T(\rho)=\phi_{t_0+T}\Xi_h(\rho,\zeta,\nu) =(p,s,n,x,y)(\rho),\qquad X_T(\rho)=(n,s,x,y)(\rho). \label{eq:delayed-projection-state} \end{equation} All primes in this subsection denote ordinary derivatives in $\rho$ with $h,a,q,D,\zeta,\nu$ fixed. In particular, the two times \[ t_0=\frac q2\log\frac1a,\qquad T=\frac Da\log\frac1a \] are constant along the family. \begin{proposition} \label{prop:delayed-projection-endpoint} There are constants $C\geq32$ and $h_*>0$ such that, for every $02$, $D>2$, and $P>6D+2$, the following holds for all sufficiently small $a>0$, uniformly over \eqref{eq:slow-variation-label-domain}. The endpoint \eqref{eq:delayed-projection-state} belongs to \eqref{eq:slow-decay-region}, has positive walls and positive height, and is ordinarily differentiable in $\rho$. With \begin{equation} B=13+16Cq+40D, \label{eq:delayed-projection-response-constant} \end{equation} its coordinates satisfy \begin{equation} \begin{gathered} 02$, the elementary limit $a^{r-2}(1+|\log a|)\to0$ as $a\downarrow0$ permits the common choice \begin{equation} a^q(1+|\log a|)\leq a^2, \qquad a^{1+D/2}(1+|\log a|)\leq a^2. \label{eq:delayed-projection-log-absorption} \end{equation} These inequalities already prove the two wall size bounds. At the endpoint, let $L$ be the logarithmic wall response from \eqref{eq:slow-variation-variables}, and put $K_0=4+8Cq+20D$. Since $aT=D|\log a|$ and $a\leq1$, \eqref{eq:slow-variation-actual-bounds} implies \begin{equation*} aL\leq K_0(1+|\log a|). \end{equation*} Multiplying $|n'|\leq nL$ by $a$ and using \eqref{eq:delayed-projection-wall-powers} and \eqref{eq:delayed-projection-log-absorption} gives $a|n'|\leq K_0a^2$. The same argument applies to $y$. Thus $|n'|,|y'|\leq K_0a\leq Ba$. For $x$, the regional bound $x\leq a^2$ gives \[ |x'|\leq xL\leq K_0a(1+|\log a|)\leq2K_0, \] where $a|\log a|\leq1$ was used in the last step. Finally $|s'|\leq5$ by the same variation theorem. The identity $B=2K_0+5$ proves the remaining bounds. \end{proof} Let $A=1+p$ at the endpoint, and use the projective coordinates \eqref{eq:projective-coordinates}. Define \begin{equation} \begin{gathered} M=\frac{\sqrt3s}{A},\qquad V=\frac{\sqrt3x}{A},\qquad k=\sqrt{(M-1)^2+V^2},\\ f_{\mathrm p}=\frac2k,\qquad g=\frac{\sqrt{s^2+(x-y)^2}}{\sqrt3}. \end{gathered} \label{eq:delayed-projection-phases} \end{equation} Here $g$ is the physical phase already controlled in Corollary~\ref{cor:small-section-slow-physical-phase}. The projection error admits the same scaled expression as in \eqref{eq:projection-scaled-error}. We use maximum norms and their induced operator norms. \begin{proposition} \label{prop:delayed-projection-error} There is an absolute constant $C_{\mathrm e}\geq1$ such that, after reducing the scale threshold in Proposition~\ref{prop:delayed-projection-endpoint}, the error $e=f_{\mathrm p}-g$ has an ordinary derivative and \begin{equation} |e(\rho)|\leq C_{\mathrm e}a^2,\qquad |e'(\rho)|\leq2C_{\mathrm e}Ba \quad(a\leq\rho\leq2a). \label{eq:delayed-projection-error} \end{equation} The constants $C_{\mathrm e}$ and the additional scale restriction are independent of $h,q,D,P$. \end{proposition} \begin{proof} Use the scaled function $F(u,z)$ from \eqref{eq:projection-rescaling}, now with the fixed compact set \[ \mathcal K=\{z:\|z\|\leq4,\ |z_1|\geq1/2\}. \] The proof of \eqref{eq:projection-scaled-bounds} gives constants $\delta>0$ and $C_{\mathrm e}\geq1$ on this set. In particular, the function is smooth near every $(u,z)$ with $\|u\|\leq\delta$ and $z\in\mathcal K$, and \begin{equation} \begin{split} |F(u,z)|&\leq C_{\mathrm e}\|u\|,\\ |DF(u,z)(u',z')| &\leq C_{\mathrm e}\|u'\| +C_{\mathrm e}\|u\|\|z'\|. \end{split} \label{eq:delayed-projection-compact-bounds} \end{equation} The second term retains $\|u\|$ because $F(0,z)$ vanishes on the entire limiting slice. At the actual endpoint take \[ u(\rho)=\left(a,\frac{n(\rho)}a,\frac{y(\rho)}a\right), \qquad z(\rho)=\left(\frac{s(\rho)}a,\frac{x(\rho)}a\right). \] The regional bounds $a/2\leq s\leq3a$ and $x\leq a^2$ place $z$ in $\mathcal K$, while \eqref{eq:delayed-projection-endpoint-bounds} gives \begin{equation*} \|u\|\leq a,\qquad \|u'\|\leq B, \qquad \|z'\|\leq\frac Ba. \end{equation*} Choose $a\leq\delta$. Since $a$ is fixed under differentiation, \eqref{eq:projection-scaled-error} and \eqref{eq:delayed-projection-compact-bounds} yield \[ \left|\mathcal P(X_T)-g\right|\leq C_{\mathrm e}a^2, \qquad \left|\frac{\dd}{\dd\rho} (\mathcal P(X_T)-g)\right| \leq aC_{\mathrm e}\left(B+a\frac Ba\right) =2C_{\mathrm e}Ba. \] It remains to identify this smooth expression with the actual projective phase also in its derivative. At each initial point the chart radicand is positive, the total time $t_0+T$ is nonnegative, and the endpoint has $p<0$ and $1+p>0$. The three strict inequalities persist for nearby chart parameters. The constraint continues to hold for their actual flow values, so \eqref{eq:projection-phase-identity} gives $\mathcal P(X_T)=f_{\mathrm p}$ on an ordinary neighborhood in those parameters. The scaled expression is smooth on a neighborhood as well. Hence this identity transfers both the value and the ordinary derivative bounds to $e$, including at the endpoints of $[a,2a]$. \end{proof} \begin{theorem} \label{thm:delayed-projection-phase} There exists $h_*>0$ such that, for every $02$, and $D>2$, one can choose $K>0$ with the following property. For every $P>6D+2$ and all sufficiently small $a>0$, uniformly over \eqref{eq:slow-variation-label-domain}, the actual phase $f_{\mathrm p}$ in \eqref{eq:delayed-projection-phases} has an ordinary derivative and satisfies \begin{equation} \begin{split} |f_{\mathrm p}(\rho)-\Phi_h(\rho,\zeta,\nu)| &\leq Ka^2,\\ \left|f_{\mathrm p}'(\rho) -\frac{\dd}{\dd\rho}\Phi_h(\rho,\zeta,\nu)\right| &\leq Ka,\\ \frac1{4\sqrt3}\leq f_{\mathrm p}'(\rho) &\leq\frac2{\sqrt3}+1. \end{split} \label{eq:delayed-projection-reference-comparison} \end{equation} The constant $K$ is chosen before $P$ and the weak-wall labels; the scale threshold may depend on $h,q,D,P$. \end{theorem} \begin{proof} Choose $h_*$ to meet the preceding propositions and Corollary~\ref{cor:small-section-slow-physical-phase}, and require $h_*\leq1/16$. Let $K_{\mathrm r}$ be the constant in that corollary, and put \[ K_{\mathrm p}=2C_{\mathrm e}B,\qquad K=K_{\mathrm p}+K_{\mathrm r}. \] These constants are selected after $h,q,D$ and before $P$. Since $B\geq1$, Proposition~\ref{prop:delayed-projection-error} bounds both projection errors by $K_{\mathrm p}a^2$ and $K_{\mathrm p}a$, respectively. The identity \[ f_{\mathrm p}-\Phi_h=e+(g-\Phi_h) \] and its ordinary derivative prove the first two lines of \eqref{eq:delayed-projection-reference-comparison}. For the derivative bounds, compare directly with the original reference phase. By Proposition~\ref{prop:small-section-reference-slope}, \[ \frac1{2\sqrt3} \leq\frac{\dd}{\dd\rho}\Phi_h(\rho,\zeta,\nu) \leq\frac2{\sqrt3}. \] Decrease the common scale threshold so that $Ka\leq1/(4\sqrt3)\leq1$. Combining these bounds with the derivative error gives the last line of \eqref{eq:delayed-projection-reference-comparison}. \end{proof} The extra waiting time has therefore preserved the reference phase through the subsequent projection. The projected state still has $V<1$ by \eqref{eq:slow-decay-actual-height}; its further approach to the growing core starts from this controlled phase.