\subsection{Comparison with the power kernel} For $u\leq7/8$ we use the auxiliary adjacent root and move it together with the kernel. The near-unit branch instead compares the augmented comparison functional at fixed geometry. Define \begin{align*} q_A(u,m)={}&\int w_\theta(s) [b_0(us)\beta_1(s)+a_0(us)\beta_m(s)]\dd s \\&-umU(u)+(u-m)(k(u)+S_\theta). \end{align*} For a configuration we will use $q_A(u,m)\geq0$. Its derivative in $m$ is strictly negative, and \[ q_A(u,u)=-\int w_\theta(s)b_0(us) [\beta_u(s)-\beta_1(s)]\dd s<0. \] For sufficiently small $m>0$, the positive term $(u-m)k(u)$ alone dominates $umU(u)$. Thus there is a unique root $m_A\in(0,u)$. For every configuration geometry $v\geq m$ and $r\leq v$, positive head and boundary-tail sampling give \[ H_{A_u}(v,r)\leq H_{A_u}(v,v)=T_{A_u}(0,v) \leq T_{A_u}(0,m)=H_{A_u}(m,m). \] Thus $q_A(u,m)\geq q(v,m,r)\geq0$ directly from the equations at the selected bodies, and strict mass decrease implies $m\leq m_A$. On $G_u(u)>0$, the identical sampling chain for the clipped kernel gives $H_{J_u}(u+v,r)\leq T_{J_u}(u,m)$. Adding the first atom and increasing head sampling therefore bound the comparison functional by $\mathcal B_u+H_{J_u}(u,m_A)$, for inner and outer geometries alike. If $G_u(u)\leq0$, the original negative baseline and one-crossing suffice. Fix $p,u,L$ as $\theta$ runs from the power law at $0$ to the circle law at $1$. The root $m_A(\theta)$ moves according to $q_A(u,m_A;\theta)=0$. Define the density deficit \[ e(s)=\rho_p(s)-w_{p,L}(s),\qquad \partial_\theta w_\theta=-e,\qquad S=\int_0^\infty e(s)\dd s. \] The last identity follows from \eqref{low:abel} by monotone convergence, since $e\geq0$. The ratio $e/w_\theta$ decreases because $w_{p,L}/\rho_p$ increases by Lemma~\ref{low:gaussianorders} for $p>1$ and by the tanh formula for $p=1$. \begin{lemma}[The needed adjacent-root bounds]\label{mid:adjacentbounds} For $1\leq p\leq9/5$ and every law in \eqref{mid:law}, \[ \max(0,2u-1)1/2$ needs proof beyond the root construction above. Put $\delta=1-u$ and $m=2u-1$. Convexity of $\beta$ gives \[ \beta_m(s)-\beta_u(s)>\delta C(us),\qquad \beta_u(s)-\beta_1(s)<\delta C(us). \] Use the same frequency function $N$ whose complete negative moment was proved in Lemma~\ref{mid:fourthroot}. Its exact identity is \[ (1+x)e^{-x}+[a_0(x)-b_0(x)]C(x)=-N(x). \] Subtract the adjacent diagonal value from its value at $m$ and use the displayed bounds. Including the self term gives \[ q_A(u,2u-1)> \delta\left\{S_\theta-\int_0^\infty w_\theta(s)N(us)\dd s\right\}>0. \] Strict decrease of $q_A$ proves the lower bound. For the pure law the quantity in braces equals $u^{-p}(Q-2P_m-1)$, where $P_m=p(p-1)\zeta(p+1)$ and $Q=p(p+1)\zeta(p+2)$. We use this power-law value in Lemma~\ref{mid:jointstrip}. \end{proof} Let \[ \mathcal L_A(u;w_\theta)=\mathcal B_u+ m_A\sum_{j\geq0}[G_u(u+jm_A)]_+. \] The next result allows its sign to be checked with the power law. \begin{lemma}[Adjacent-root comparison]\label{mid:adjacentreturn} If $\mathcal L_A(u;\rho_p)<0$, then the comparison functional is strictly negative for every configuration bounded by this adjacent-root mesh. \end{lemma} \begin{proof} If $G_u(u)\leq0$, the one-crossing property makes every following positive part vanish. Moreover \[ Z-k(1)-aU(1)=\int w_\theta(s)e^{-s} \left[\frac{s}{1-e^{-s}}-1-as\right]\dd s>0, \] since $s/(1-e^{-s})>1+s/2$ and $a\leq6/25$. Hence $\mathcal B_u<0$. It remains to consider portions of the law path where $G_u(u)>0$. Apply the two sampling monotonicities of Lemma~\ref{low:packing} to $J=[G_u]_+$: since $H_J= mJ(u)+T_J$, they give \[ 0\leq\partial_m\left(m\sum_{j\geq0}[G_u(u+jm)]_+\right) \leq G_u(u). \] This includes one-sided derivatives at a clipping contact. Set $\delta=1-u$, $X=e^{-s}$, $Y=e^{-us}$ and \[ A_0=1-(1+as)X,\quad \gamma=A_0-zs,\quad B=\beta_u(s)-\beta_1(s),\quad D=u+\beta_u(s)-m-\beta_m(s). \] Let $\mathcal L_A^{\rm sgn}$ denote the comparison functional with the full signed right mesh. The geometric sums give its density $u[X+asX-sX/(1-X)]-Y(m+\beta_m(s))\gamma$. Adding $\delta$ for the self term and simplifying gives $R_L$ below. For $q_A$, combine its diagonal value at $m=u$ with the change from $u$ to $m$. Thus the fixed-mass derivatives of $\mathcal L_A^{\rm sgn}$ and $q_A$ are $-\int e R_L\dd s$ and $-\int e R_q\dd s$, respectively, where \begin{equation}\label{mid:returnidentities} R_L=\delta A_0+\gamma(YD-B),\qquad -R_q=b_0(us)B+a_0(us)D. \end{equation} For clarity, write $D_A=-q_{A,m}>0$ and use a partial $\theta$ derivative to mean that the mass is held fixed. Then \[ 0=\frac{\dd}{\dd\theta}q_A(m_A(\theta),\theta) =-\int eR_q\dd s-D_A m_A',\qquad m_A'=-\frac{\int eR_q\dd s}{D_A}. \] The fixed-mass comparison functional derivative is at most $-\int eR_L\dd s$. Adding the chain-rule term $(\partial_m\mathcal L_A)m_A'$ therefore gives, almost everywhere, \[ \frac{\dd}{\dd\theta}\mathcal L_A \leq-\int e(s)[R_L+\lambda R_q]\dd s, \quad \lambda=\frac{\partial_m\mathcal L_A}{-q_{A,m}}. \] At each site where $G_u(t)\leq0$, its negative-to-positive density crossing and the decreasing weight $e/w_\theta$ give $\int e(s)e^{-ts}sg_u(s)\dd s\leq0$. The signed site therefore has nonnegative fixed-mass law derivative. Clipping it to zero can only decrease that derivative, which gives the inequality above. Since $u<1$, both endpoint kernels are negative beyond a common finite distance, and so are their convex combinations. The clipped sum is locally finite along the positive root path. For the signed sum the power-law bound on the tail justifies termwise differentiation. The kernel elasticity and $w_\theta(s)/s$ decreasing give, with $R=u/(1+u)$, \[ 0\leq\lambda\leq\frac{G_u(u)}{uU(u)+k(u)+S_\theta} \leq\frac{1+R^2-2(a/u)(1-R^3)}3<\frac{83}{300}. \] Here is a direct signed-moment proof of the middle inequality. After dropping the positive self term from the denominator, the ratio is the quotient of the integrals with densities $w_\theta e^{-us}sg_u(s)$ and $w_\theta e^{-us}(us+1)$. On the positive part of $g_u$, the function $sg_u(s)/(us+1)$ strictly increases, since both factors are positive increasing. For a positive constant $c$, the difference $c(us+1)-sg_u(s)$ therefore has at most one crossing, from positive to negative. Choose $c$ to make its integral against $se^{-us}$ zero. Using $\int_0^\infty s^je^{-ts}\dd s=j!t^{-j-1}$ gives exactly $c=[1+R^2-2(a/u)(1-R^3)]/3$. Multiplication by the decreasing weight $w_\theta(s)/s$ makes the integral nonnegative, proving the ratio bound. The resulting rational expression increases with $u$ and equals $83/300$ at $u=1$. Also, with $x=us$ and $E(x)=1-C(x)$, the root bound gives \[ B=\int_u^1C(ts)\dd t\leq\delta C(x),\qquad D=\int_m^uE(ts)\dd t<\delta E(x). \] Here $C$ decreases, $E$ increases, and $u-m<\delta$. We keep these two bounds together. If $\gamma\geq0$, then \[ R_L+\lambda R_q \geq\delta\{A_0-\gamma C(x) -\lambda[b_0(x)C(x)+a_0(x)E(x)]\} \geq\delta\{(z-\lambda u)s+\gamma E(x)\}>0, \] since $083/300>\lambda$. If $\gamma<0$, put $g=-\gamma>0$. Dropping $gB>0$ and using $B\leq\delta C(x)$, $D<\delta$, gives \[ R_L+\lambda R_q\geq\delta\{A_0-Yg-\lambda K(x)\}, \qquad K(x)=a_0(x)+b_0(x)C(x). \] Write $\sigma=a/u\leq12/25$. Since $X\leq Y=e^{-x}$, \[ \frac{A_0-Yg-(1-Y)/2}{Y} \geq\sinh x-x+\frac{1-e^{-x}}2-\sigma xe^{-x}>0. \] The last sign uses $\sinh x>x$ and $e^x>1+x$. For the other term put $T=h_0(x)-1$. The common envelope \eqref{pf:bose} gives $Te^{-x}\left\{\frac{e^x-1}{2}+x -\frac{x^2}4-\frac{x^3}{12}-\frac{x^4}{144}\right\} >e^{-x}\left(\frac{3x}2+\frac{x^4}{72}\right)>0. \] Consequently $R_L+\lambda R_q>(17/200)\delta(1-Y)>0$. The comparison functional therefore decreases on every active portion of the law path. At a boundary of such a portion it is already negative by the first paragraph. This proves the implication from the power-law value. \end{proof}