\section{Preliminaries}\label{sec:preliminaries} Throughout the paper the offspring variable satisfies $\PP(\nu\ge1)=1$ and $11$, we establish the kernel identities and linear-response formula used in the two proofs in Section~\ref{sec:twochildren} and in the later estimates. \subsection{The escape law and the speed formula} Attach an artificial parent to the root of a descendant tree $T$, and let $\beta_\lambda(T)$ be the quenched probability, starting from that root, of never visiting its parent. For $0<\lambda1$, then $\lambda-1\le A_\lambda\le\lambda$ on $[0,1]^2$. If $0<\lambda<1$, the finite recursion gives $X\ge1-\lambda$ under $\pi_\lambda$, and \begin{equation}\label{eq:low-denominator} A_\lambda(x,y)\ge\lambda(1-\lambda)>0 \quad\text{for }x,y\in[1-\lambda,1]. \end{equation} \begin{proposition}[Symmetric speed formula]\label{prop:symmetric} For $0<\lambda0, \qquad 0<\lambda<\Lambda. \end{equation*} Consequently, whenever $|\lambda-1|<\beta_\Lambda(T)/2$, \begin{equation*} \left| \frac{\sum_{i=0}^{\nu}X_i} {\lambda-1+\sum_{i=0}^{\nu}X_i}-1 \right| \le\frac{2|\lambda-1|}{\beta_\Lambda(T)} \longrightarrow0. \end{equation*} We next bound the integrands in~\eqref{eq:speed-exchangeable} uniformly near one. For $\lambda\ge1$, \begin{equation*} 0<\frac{\sum_{i=0}^{\nu}X_i} {\lambda-1+\sum_{i=0}^{\nu}X_i}\le1. \end{equation*} For $0<\lambda<1$, the escape recursion with terminal value one gives $X_i\ge1-\lambda$. Indeed, if every child value is at least $1-\lambda$, then their sum $S$ is at least $1-\lambda$, and \begin{equation*} \frac{S}{\lambda+S} \ge\frac{1-\lambda}{\lambda+1-\lambda}=1-\lambda. \end{equation*} Induction followed by the limit in depth proves the bound. Since $\nu\ge1$, it follows that \begin{equation*} 0<\frac{\sum_{i=0}^{\nu}X_i} {\lambda-1+\sum_{i=0}^{\nu}X_i} =1+\frac{1-\lambda}{\sum_{i=0}^{\nu}X_i-(1-\lambda)} \le1+\frac1\nu\le2. \end{equation*} Also, for $0<\lambda<\Lambda$, \begin{equation*} \frac{|\nu-\lambda|}{\nu+1}\le1+\Lambda, \qquad \frac{\nu+\lambda}{\nu+1}\le1+\Lambda. \end{equation*} Thus both integrands in~\eqref{eq:speed-exchangeable} are bounded in absolute value by $2(1+\Lambda)$. Dominated convergence shows that the numerator tends to $\E[(\nu-1)/(\nu+1)]$ and the denominator tends to one. Evaluating~\eqref{eq:speed-exchangeable} at $\lambda=1$ gives \begin{equation*} \lim_{\lambda\to1}v(\lambda) =\E\left[\frac{\nu-1}{\nu+1}\right] =v(1).\qedhere \end{equation*} \end{proof} \subsection{Continuity of the escape law and speed for \texorpdfstring{$1<\lambda0. \end{equation} To compare two ordered fixed laws, couple $X_0\le X_1$ with laws $\rho,\sigma$ and interpolate $X_s=(1-s)X_0+sX_1$. For the recursion residual \begin{equation*} \res_{u_\rho}(\lambda,\sigma) =\E_\sigma u_\rho(X) -\E_{\nu,\sigma}u_\rho\left(F_\lambda\left(\sum_{i=1}^{\nu}X_i\right)\right), \end{equation*} the chain rule along independent copies of this coupling yields \begin{align*} \frac{\dd}{\dd s}\res_{u_\rho}(\lambda,\Law(X_s)) &=\E\left[(X_1-X_0) \{r_\rho-\cL_{\lambda,\Law(X_s)}r_\rho\}(X_s)\right]\\ &\ge\frac{\E_\rho Y^2}{\lambda^2}\E(X_1-X_0). \end{align*} For a fixed number of children, this is the ordinary chain rule: the derivative of the sum is the sum of the coupled increments, and their contributions agree by exchangeability. The recursion term is bounded by $\nu\|r_\rho\|_\infty/\lambda$, since each increment is at most one; the direct term is bounded by $\|r_\rho\|_\infty$. These bounds justify differentiation under the offspring expectation. The residual vanishes at both $s=0$ and $s=1$, because both endpoint laws satisfy the recursion. Integration in $s$ therefore forces $\E(X_1-X_0)=0$ and hence $\rho=\sigma$. Iteration from terminal value one dominates every fixed law: couple its terminal ones above independent samples of that fixed law and iterate. Its limit has law $\pi_\lambda$, so the preceding ordered comparison proves uniqueness of the nonzero fixed law. At any $\lambda\in(1,m)$, each one-sided same-tree limit of $\beta_\alpha$ as $\alpha\to\lambda$ satisfies~\eqref{eq:rde}. This follows by bounded convergence, since the number of children is finite almost surely. Choose $\Lambda\in(\lambda,m)$. The one-sided limits dominate $\beta_\Lambda>0$, and hence their laws are nonzero. Uniqueness identifies both with $\pi_\lambda$. The order~\eqref{eq:same-tree} then gives \begin{equation*} W_1(\pi_\alpha,\pi_\lambda) =\E|\beta_\alpha-\beta_\lambda|\longrightarrow0. \end{equation*} The kernels in~\eqref{eq:NZ} are bounded and continuous for biases near $\lambda>1$. Since $Z_\lambda(\pi_\lambda)>0$, the speed formula gives continuity of $v$. \end{proof} \subsection{Kernel identities and the weighted mean}\label{sec:kernel-identities} Fix $1<\lambda0,\qquad K>0. \end{equation*} \end{lemma} \begin{proof} We first show that $\cJ h_v$ is uniformly positive. Put $W=\E[X/A(X,Y)]$. Symmetry gives $N+Z=2W$. For fixed $x$, the function $y\mapsto A(x,y)^{-1}$ is decreasing; its covariance with $Y$ is nonpositive. Hence $N\le(\E X)W$ and \begin{equation}\label{eq:weighted-speed} \frac{v}{1+v}=\frac{N}{2W}\le\frac{\E X}{2}0, \qquad 0\le x\le1. \end{equation} By~\eqref{eq:positive-K}, this positivity and $00$. \end{proof} \input{sections/02_response}