\documentclass[11pt,letterpaper]{article} \pdfgentounicode=1 \input{glyphtounicode} \pdfglyphtounicode{parenleftbig}{0028} \pdfglyphtounicode{parenrightbig}{0029} \pdfglyphtounicode{bracketleftbig}{005B} \pdfglyphtounicode{bracketrightbig}{005D} \pdfglyphtounicode{braceleftbig}{007B} \pdfglyphtounicode{bracerightbig}{007D} \pdfglyphtounicode{floorleftbig}{230A} \pdfglyphtounicode{floorrightbig}{230B} \pdfglyphtounicode{ceilingleftbig}{2308} \pdfglyphtounicode{ceilingrightbig}{2309} \pdfglyphtounicode{parenleftBig}{0028} \pdfglyphtounicode{parenrightBig}{0029} \pdfglyphtounicode{bracketleftBig}{005B} \pdfglyphtounicode{bracketrightBig}{005D} \pdfglyphtounicode{braceleftBig}{007B} \pdfglyphtounicode{bracerightBig}{007D} \pdfglyphtounicode{floorleftBig}{230A} \pdfglyphtounicode{floorrightBig}{230B} \pdfglyphtounicode{ceilingleftBig}{2308} \pdfglyphtounicode{ceilingrightBig}{2309} \pdfglyphtounicode{vextendsingle}{2223} \pdfglyphtounicode{vextenddouble}{2225} \pdfglyphtounicode{summationtext}{2211} \pdfglyphtounicode{summationdisplay}{2211} \pdfglyphtounicode{uniontext}{22C3} \pdfglyphtounicode{uniondisplay}{22C3} \pdfglyphtounicode{integraltext}{222B} \pdfglyphtounicode{integraldisplay}{222B} \pdfglyphtounicode{bracelefttp}{23A7} \pdfglyphtounicode{braceleftmid}{23A8} \pdfglyphtounicode{braceleftbt}{23A9} \pdfglyphtounicode{braceex}{23AA} \pdfglyphtounicode{bar}{2223} \pdfglyphtounicode{backslash}{2216} \pdfglyphtounicode{openbullet}{2218} \pdfglyphtounicode{periodcentered}{22C5} \pdfglyphtounicode{mapsto}{21A6} \pdfglyphtounicode{Delta}{0394} \pdfglyphtounicode{tilde}{007E} \pdfglyphtounicode{circumflex}{005E} \pdfglyphtounicode{macron}{02C9} \usepackage[T1]{fontenc} \usepackage[utf8]{inputenc} \renewcommand{\rmdefault}{lmr} \renewcommand{\sfdefault}{lmss} \renewcommand{\ttdefault}{lmtt} \usepackage[top=0.9in,bottom=0.9in,left=0.95in,right=0.95in,footskip=0.45in]{geometry} \usepackage{amsmath} \usepackage{amssymb} \usepackage{graphicx} \usepackage{array} \usepackage{longtable} \usepackage{booktabs} \usepackage{ragged2e} \usepackage{enumitem} \usepackage{url} \usepackage[hidelinks]{hyperref} \hypersetup{pdftitle={Essential enhancements of site percolation in every dimension and on periodic lattices},pdfauthor={Tejasvi Singh Tomar},pdfsubject={Site percolation; essential enhancements; Aizenman-Grimmett strict inequality; periodic lattices}} \urlstyle{same} \DeclareUnicodeCharacter{2212}{\textminus} \DeclareUnicodeCharacter{2032}{\ensuremath{'}} \linespread{1.1} \setlength{\parindent}{0pt} \setlength{\parskip}{5pt plus 1pt minus 1pt} \frenchspacing \tolerance=10000 \setlength{\emergencystretch}{3em} \hbadness=10000 \makeatletter \renewcommand\section{\@startsection{section}{1}{\z@}{16pt plus 4pt minus 2pt}{6pt plus 2pt}{\normalfont\large\bfseries}} \renewcommand\subsection{\@startsection{subsection}{2}{\z@}{12pt plus 3pt minus 2pt}{4pt plus 1pt}{\normalfont\normalsize\bfseries}} \renewcommand\subsubsection{\@startsection{subsubsection}{3}{\z@}{10pt plus 3pt minus 2pt}{4pt plus 1pt}{\normalfont\normalsize\bfseries}} \newcommand{\cmrtext}{\fontencoding{OT1}\fontfamily{cmr}\fontseries{m}\fontshape{n}\selectfont} \DeclareRobustCommand{\text}{\ifmmode\expandafter\mdtext@\else\expandafter\mbox\fi} \def\mdtext@#1{\text@{\cmrtext #1}} \def\tagform@#1{\maketag@@@{\cmrtext(\ignorespaces#1\unskip\@@italiccorr)}} \makeatother \newcommand{\mdtitle}[1]{\begin{center}{\fontsize{17}{21}\selectfont\bfseries #1\par}\end{center}\vspace{-4pt}} \newcommand{\qedsq}{{\unskip\nobreak\hfil\penalty50\hskip1em\hbox{}\nobreak\hfil$\square$\parfillskip=0pt\finalhyphendemerits=0\par}} \newcommand{\discbullet}{\ifdefined\pdfliteral\leavevmode\hbox to 4.2pt{\pdfliteral{q 0 g 3.8 2.2 m 3.8 3.194 2.994 4 2 4 c 1.006 4 0.2 3.194 0.2 2.2 c 0.2 1.206 1.006 0.4 2 0.4 c 2.994 0.4 3.8 1.206 3.8 2.2 c f Q}\hss}\else\textbullet\fi} \setlist{topsep=4pt,partopsep=0pt,itemsep=2pt,parsep=0pt} \setlist[itemize]{label=\discbullet,leftmargin=22pt} \setlist[enumerate]{label=\arabic*.,leftmargin=22pt} \newcolumntype{P}[1]{>{\RaggedRight\tolerance=10000\arraybackslash}p{#1}} \setlength{\LTleft}{0pt} \setlength{\LTright}{0pt plus 1fill} \setlength{\LTpre}{6pt} \setlength{\LTpost}{6pt} \pagestyle{plain} \begin{document} \mdtitle{Essential enhancements of site percolation in every dimension and on periodic lattices} \begin{center} \textbf{Tejasvi Singh Tomar} \end{center} \begin{center} Version 6, 27 October 2026 \end{center} \section*{Abstract} We prove Conjecture 1 of Balister, Bollobás and Riordan for site percolation on $\mathbb Z^d$ in every dimension $d\ge2$: for every essential enhancement, activated at any positive density, the enhanced configuration percolates for all $p0$. Then for any $s>0$ there is a $\pi(s)0$ for all $p$ satisfying $\pi(s)0$ there is $\pi(s)0$ for all $p$ with $\pi(s)0$ for all $p$ with $\pi(s)0$ for all $p$ with $\pi(s)0$ for all $p$ with $\pi(s)r_E$ and a finite set $\omega_g\subseteq\mathbb Z^d$ such that} \begin{enumerate} \item{} \textit{$\omega_g\subseteq B_m$ and $\omega_g\cap S_m=\{me_1,-me_1\}$;} \item{} \textit{$\omega_g$ contains no path from $-me_1$ to $me_1$;} \item{} \textit{$\omega_g\cup E_0(\omega_g)$ contains a path from $-me_1$ to $me_1$.} \end{enumerate} \textit{Moreover $E_0(\omega_g)\subseteq B_{r_E}\subseteq B_{m-1}$.} \textit{Proof.} Balister, Bollobás and Riordan prove the $\ell^\infty$ form [BBR, §4] and note that either form implies the other. We derive the $\ell^1$ form. Write $B^\infty_M$ and $S^\infty_M$ for $\ell^\infty$ balls and spheres. Since $B_{r_E}\subseteq B^\infty_{r_E}$, the map $E_0$ also has range $r_E$ with respect to the $\ell^\infty$ metric. The $\ell^\infty$ form of the lemma gives an integer $M>r_E$ and a finite set $\omega_\infty\subseteq B^\infty_M$ with $\omega_\infty\cap S^\infty_M=\{\pm Me_1\}$, such that $\omega_\infty$ contains no path from $-Me_1$ to $Me_1$ while $\omega_\infty\cup E_0(\omega_\infty)$ does. Put $m=dM$, let $\Sigma_\pm=\{\pm te_1: MM>r_E$. (2) A site $te_1$ with $t>M$ has the neighbours $(t\pm1)e_1$ and $te_1\pm e_j$ ($j\ge 2$). The latter have $\ell^\infty$ norm $t>M$, so they are not in $\omega_\infty$, and $(t-1)e_1\in\omega_\infty$ only if $t=M+1$. Thus $\Sigma_+$ is a path attached to $\omega_\infty\cup\Sigma_-$ only through the edge between $(M+1)e_1$ and $Me_1$, and similarly for $\Sigma_-$. A path in $\omega_g$ from $-me_1$ to $me_1$ would therefore contain a path in $\omega_\infty$ from $-Me_1$ to $Me_1$, which does not exist. (3) $E_0(\omega_g)$ depends only on $\omega_g\cap B_{r_E}$. Since $B_{r_E}\subseteq B^\infty_{M-1}$ and $\Sigma_\pm\cap B^\infty_M=\emptyset$, we have $\omega_g\cap B_{r_E}=\omega_\infty\cap B_{r_E}$, so $E_0(\omega_g)=E_0(\omega_\infty)\subseteq B_{r_E}$. A path in $\omega_\infty\cup E_0(\omega_\infty)$ from $-Me_1$ to $Me_1$, extended along $\Sigma_-$ and $\Sigma_+$, joins $-me_1$ to $me_1$ in $\omega_g\cup E_0(\omega_g)$.\qedsq From now on $E_0$, $r_E$, $m$ and $\omega_g$ are fixed as in Lemma 2.2. Note that $E_0(\omega_g)\subseteq B_{m-1}$: activating the enhancement at the centre of a copy of the gadget changes nothing on or outside $S_m$, and it joins the two terminals $\pm me_1$, which $\omega_g$ alone does not join. \subsection*{2.4. Two consecutive layers} We need the following elementary fact in Lemma 3.1 and in the boundary case of Theorem 2.13. \textbf{Lemma 2.3.} \textit{Let $d\ge2$, $n\ge1$, $i\in\{1,\dots,d\}$, and let $ss\}$ contains $ne_i$ and is connected. In particular $S_n\cup S_{n+1}$ is connected.} \textit{Proof.} Let $x\in A$. We walk from $x$ to $ne_i$ inside $S_n\cup S_{n+1}$, changing one coordinate by one unit at each step. Write $y$ for the current site. \textit{Phase 1, while $y_i<0$.} If $|y|=n$, move some coordinate $y_j$ with $j\ne i$ one unit away from $0$ (this is possible since $d\ge2$). The new site lies in $S_{n+1}$. If $|y|=n+1$, replace $y_i$ by $y_i+1$. As $y_i<0$, the new site lies in $S_n$. Every two steps raise $y_i$ by one, so eventually $y_i=0$. \textit{Phase 2, while $y_i\ge0$.} If $y=ne_i$, stop. If $y=(n+1)e_i$, step to $ne_i$ and stop. If $|y|=n+1$ and $y\ne(n+1)e_i$, some coordinate $y_j$ with $j\ne i$ is nonzero. Move it one unit towards $0$, which leads to $S_n$. If $|y|=n$ and $y\ne ne_i$, replace $y_i$ by $y_i+1$, which leads to $S_{n+1}$. Two consecutive steps starting in $S_n$ raise $y_i$ by one and lower $\sum_{j\ne i}|y_j|=n-y_i$ by one, so the walk reaches $ne_i$. Along the walk $y_i$ never decreases, except in a final step from $(n+1)e_i$ to $ne_i$. Since $x_i>s$ and $n>s$, the walk stays in $A$. For the last assertion take $s=-n-2$.\qedsq \subsection*{2.5. The deletion step} In this subsection and the next we fix integers $K\ge m+1$ and $L$, and a state $(\omega,\alpha)$ in which a site $v$ is p-pivotal for $R_L$, such that \[ K+10<|v|K+r_E$, so $B_{r_E}(a)\cap\Omega=\emptyset$. The set $E_0(\tilde\omega-a)+a$ depends only on $\tilde\omega\cap B_{r_E}(a)$, where $\tilde\omega$ agrees with $\omega$, and it lies in $B_{r_E}(a)$. Hence $Y=\bigcup_{a\in\alpha}(E_0(\omega-a)+a)$ has the required properties. Finally $c\notin\alpha$ because $\Omega\subseteq B_{K+r_E}(v)$.\qedsq \textbf{Lemma 2.5} (the deletion step; [BBR, p. 9]). \textit{There are a configuration $\omega'$ that agrees with $\omega$ outside $\Omega$ and an induced path $P=(p_0,\dots,p_n)$ with $p_0=0$, $p_n\in S_L$ and $|p_j|K+10$ and $|p_n-v|\ge L-|v|>K+10$.\qedsq From now on $\omega'$ and $P$ are as in Lemma 2.5. We order $P$ from $p_0=0$ to $z:=p_n$, and for $x\in P\setminus\{0,z\}$ we write $\mathrm{pred}(x)$ and $\mathrm{succ}(x)$ for the sites before and after $x$ on $P$. \textbf{Lemma 2.6} (the class property; [BBR, pp. 10–11]). \textit{Let $D\subseteq\Omega$ meet $P$. Write $P=O_0I_1O_1\cdots I_kO_k$ with $k\ge1$, where $I_1,\dots,I_k$ are the maximal runs of consecutive sites of $P$ in $D$ and $O_0,\dots,O_k$ are the runs between them. ($O_0\ni0$ and $O_k\ni z$ are nonempty because $0,z\notin\Omega$.) Put $O_D=E(\omega',\alpha)\setminus D$, and let $H_j$ be the component of $O_D$ that contains $O_j$. Then $H_0,\dots,H_k$ are pairwise distinct, and $H_j\cap S_L=\emptyset$ for every $jm+1>r_E+1$. \textbf{Proposition 2.12.} \textit{Assume CL($d$) and define $\rho_0$ and $K$ by (2.2). Let $(\omega,\alpha)$ be a state in which $v$ is p-pivotal for $R_L$, and suppose that $K+10<|v|2K+90$, so $B_{r_E}(a)$ misses $B_{K+21}$. As in Lemma 2.4, there is a set $Y'$ disjoint from $B_{K+21}$ such that $E(\hat\omega,\alpha)=\hat\omega\cup Y'$ for every $\hat\omega$ that agrees with $\omega$ outside $B_{K+21}$. The configuration $E(\tilde\omega,\alpha)$ contains a path from $0$ to $S_L$. Indeed, $E(\omega\cup\{v\},\alpha)$ contains such a path $\Gamma$. Let $x'$ be the last site of $\Gamma$ in $B_{K+21}$, so that $|x'|=K+21$. The part of $\Gamma$ after $x'$ lies in $(\omega\cup\{v\}\cup Y')\setminus B_{K+21}=E(\tilde\omega,\alpha)\setminus B_{K+21}$. Inside $B_{K+21}$, the configuration $\tilde\omega$ contains the connected set $B_{K+18}$, which contains $0$ and an inner neighbour of $w$; it contains $w$; and it contains $S_{K+20}\cup S_{K+21}$, which is connected by Lemma 2.3 and contains $x'$ and an outer neighbour of $w$. Concatenation gives the path. On the other hand, every path from $0$ to $S_L$ meets $S_{K+19}$, and $w$ is the only site of $S_{K+19}$ in $E(\tilde\omega,\alpha)$. Therefore $w$ is p-pivotal for $R_L$ in $(\tilde\omega,\alpha)$. Moreover $K+10<|w|=K+1969$. Taking $w$ nearest to $v$ repairs this. Case 3 follows [BBR, pp. 11–12], with $r$ replaced by $K$. \section*{3. The two-port theorem in every dimension} In this section the centre is $0$. By translation, all statements hold for balls $B_r(c)$, with coordinates taken relative to $c$ and terminals $c\pm me_1$. For a site $x$ with $x_i\ne0$, the \textit{inward step} of $x$ in coordinate $i$ is $x-\operatorname{sgn}(x_i)e_i$, and it lowers the layer by one. \subsection*{3.1. Lemma 8 of [BBR] in every dimension} \textbf{Lemma 3.1} ([BBR, Lemma 8]). \textit{Let $d\ge2$, $m\ge1$ and $r\ge m+10$. Let $a,g\in S_r$ be separated, that is, for some coordinate $i$ there is an integer $s$ with $|s|\le r-5$ strictly between $a_i$ and $g_i$. Then there are a red path starting at $a$ and a green path starting at $g$ with the following properties: one of them ends at $me_1$ and the other at $-me_1$; all their sites other than $a$ and $g$ lie in $B_{r-1}$; all their sites other than these two end sites lie outside $B_m$; and no red site equals or is adjacent to a green site.} [BBR] state and prove this lemma for general $d$. One step of their proof, the connectivity of the two halves of an annulus, is asserted without proof. Lemma 2.3 supplies it. \textit{Proof.} The statement is symmetric under exchanging $a$ and $g$, so we may assume $a_i>s>g_i$. Put $w=(r-2)e_i$ and \[ A_+=(S_{r-1}\cup S_{r-2})\cap\{x:x_i>s\},\qquad A_-=(S_{r-1}\cup S_{r-2})\cap\{x:x_is\ge-(r-5)$ forces $a=re_i$, whose neighbour $(r-1)e_i$ lies in $A_+$. In the same way $g$ has a neighbour in $A_-$. \textit{First stage.} Join $a$ by a red walk through $A_+$ to $w$, and continue it with $(r-3)e_i$ and $w'=(r-4)e_i$. Join $g$ by a green walk through $A_-$ to $-w$, and continue it with $-(r-3)e_i$ and $-w'$. Every red site $x$ so far satisfies $x_i\ge s+1$, and every green site satisfies $x_i\le s-1$: this holds on $A_\pm$ by definition, at $a$ and $g$ by hypothesis, and at $\pm(r-3)e_i$ and $\pm w'$ because $|s|\le r-5$. Hence no red site equals or is adjacent to a green site. Apart from $a$ and $g$, all first-stage sites lie in layers $r-1,\dots,r-4$, and the only first-stage sites in layers $r-3$ and $r-4$ are $\pm(r-3)e_i$ and $\pm w'$. \textit{Second stage.} All further sites lie in layers at most $r-4$. Such a site can be adjacent to a first-stage site only if that site is $\pm w'$, or if it lies in layer $r-4$ and is adjacent to $\pm(r-3)e_i$. The only site of layer $r-4$ adjacent to $(r-3)e_i$ is $w'$, and similarly for $-(r-3)e_i$. So it suffices to join $w'$ and $-w'$ to the terminals by a red and a green walk without red–green contacts among themselves. If $i=1$, go along the axis from $w'=(r-4)e_1$ to $me_1$, and from $-w'$ to $-me_1$. If $i\ne1$, we may assume $i=2$ after permuting the coordinates $2,\dots,d$, which fixes $\pm me_1$. Join $w'=(r-4)e_2$ to $(r-4)e_1$ by the staircase \[ (r-4)e_2,\ (r-5)e_2,\ e_1+(r-5)e_2,\ e_1+(r-6)e_2,\ \dots,\ (r-5)e_1,\ (r-4)e_1, \] whose sites $x$ satisfy $x_1,x_2\ge0$ and $x_1+x_2\in\{r-5,r-4\}$, and then go along the axis to $me_1$. The green walk is the reflection of the red one through the origin. In both cases every red second-stage site has nonnegative coordinates and every green one has nonpositive coordinates, and all of them have norm at least $m$. Consequently a red site $x$ and a green site $y$ satisfy $|x-y|=|x|+|y|\ge2m\ge2$. \textit{Location.} First-stage sites other than $a$ and $g$ lie in layers $r-4,\dots,r-1$, and $r-4\ge m+6$. Second-stage sites lie in layers at most $r-4$. The staircase lies in layers $r-5$ and $r-4$, which exceed $m$, and the axis segments meet $B_m$ only at $\pm me_1$.\qedsq \subsection*{3.2. Escaping from a non-corner pair} \textbf{Theorem 3.2} (escape). \textit{Let $d\ge2$ and $r\ge\max(2d+1,20)$. Let $a\ne g$ be sites of $S_r$ that do not form a corner pair. Then there are $k\in\{0,1,2,3\}$ and walks $\gamma_a$ from $a$ to a site $a^*$ and $\gamma_g$ from $g$ to a site $g^*$ such that:} \begin{itemize} \item{} \textit{(E1) $a^*,g^*\in S_{r-k}$, every site of $\gamma_a$ and $\gamma_g$ other than $a$ and $g$ lies in the layers $r-k,\dots,r-1$, and layer $r-k$ is met only by $a^*$ and $g^*$;} \item{} \textit{(E2) no site of $\gamma_a$ equals or is adjacent to a site of $\gamma_g$;} \item{} \textit{(E3) $a^*$ and $g^*$ are separated in $S_{r-k}$.} \end{itemize} For $k=0$ the walks are $\gamma_a=(a)$ and $\gamma_g=(g)$. Together with Lemma 3.1, the theorem replaces Lemma 9 of [BBR], which is specific to $d=3$, by a statement valid in every dimension. Its proof distinguishes five cases. \textit{Notation.} Put $\Delta=g-a$. Since $|a|=|g|=r$, the norm $|\Delta|$ is even and at least $2$. We use the following bookkeeping rule throughout: when a site of one walk takes a step in coordinate $j$, its distance to a fixed site $y$ of the other walk grows by one if the step moves its $j$-th coordinate away from $y_j$, and drops by one otherwise. \textit{Step 0.} If $a$ and $g$ are separated in $S_r$, take $k=0$. From now on we assume that they are not. If $|\Delta_j|\ge2$ for some $j$, the integers strictly between $a_j$ and $g_j$ form a nonempty interval that avoids $[-(r-5),r-5]$. Hence $a_j$ and $g_j$ have the same sign and $|a_j|,|g_j|\ge r-5$. Since $r\ge20$, each of $a$ and $g$ has at most one coordinate of absolute value at least $r-7$. \textbf{Type II: every coordinate of $a$ and of $g$ has absolute value at most $r-7$.} By Step 0, $|\Delta_j|\le1$ for every $j$, and neither $a$ nor $g$ is a corner of $S_r$. Call $j\in\operatorname{supp}\Delta$ \textit{good} if $a_j$ and $g_j$ are both nonzero. They then have the same sign and differ by one. \textit{Case II-A: a good $j$ exists.} Let $X$ be the one of $a,g$ with the smaller $|X_j|$ and $Y$ the other, so that $Y_j=X_j+\operatorname{sgn}(X_j)$. Put $X'=X-\operatorname{sgn}(X_j)e_j$. Since $Y$ is not a corner, it has a nonzero coordinate $i\ne j$, and we put $Y'=Y-\operatorname{sgn}(Y_i)e_i$. When $|\Delta|=2$ and $\operatorname{supp}\Delta=\{j,l\}$, we choose $i\notin\{j,l\}$ if $Y$ has such a nonzero coordinate, and $i=l$ otherwise. Let $\gamma_X=(X,X')$ and $\gamma_Y=(Y,Y')$, so $k=1$. The step from $X$ to $X'$ moves $X_j$ away from $Y_j$, so $|X'-Y|=|\Delta|+1$. The sites $X'$ and $Y'$ have $j$-th coordinates $X_j-\operatorname{sgn}X_j$ and $X_j+\operatorname{sgn}X_j$, so $|X'-Y'|\ge2$. If $|\Delta|\ge4$, then $|X-Y'|\ge|\Delta|-1\ge3$. Let $|\Delta|=2$ with $\operatorname{supp}\Delta=\{j,l\}$. If $i\notin\{j,l\}$, then $X_i=Y_i\ne0$, the step moves $Y_i$ away from $X_i$, and $|X-Y'|=3$. Otherwise $\operatorname{supp}Y=\{j,l\}$ and $i=l$. If the step moved $Y_l$ towards $X_l$, we would have $|X_l|=|Y_l|-1$. Together with $|X_j|=|Y_j|-1$ and $X=Y=0$ outside $\{j,l\}$ this gives $|X|=|Y|-2$, which is false. So again $|X-Y'|=3$, and (E2) holds. Both $X'$ and $Y'$ lie in layer $r-1$, which gives (E1). For (E3), the integer $s=X_j$ lies strictly between $X'_j$ and $Y'_j=Y_j$, and $|s|\le r-7\le(r-1)-5$. \textit{Case II-B: no good $j$ exists.} Then every $j\in\operatorname{supp}\Delta$ has $\{a_j,g_j\}=\{0,\pm1\}$. If $|a_j|\le1$ also held for every $j\notin\operatorname{supp}\Delta$, we would get $r=|a|\le d$, contrary to $r\ge2d+1$. So some $j\notin\operatorname{supp}\Delta$ has $|a_j|\ge2$, and there $a_j=g_j=:x$. Put $\sigma=\operatorname{sgn}x$ and $\gamma_a=(a,\,a-\sigma e_j,\,a-2\sigma e_j)$. Let $\gamma_g=(g,g_1,g_2)$ consist of two successive inward steps of $g$ in coordinates other than $j$, chosen as follows. If $|\Delta|=2$, then $\operatorname{supp}\Delta=\{j_1,j_2\}$, where $(a,g)$ takes the values $(0,\pm1)$ in one of these coordinates and $(\pm1,0)$ in the other, because $|a|=|g|$. We take both steps in coordinates outside $\operatorname{supp}\Delta\cup\{j\}$, where $g$ has total mass $r-|x|-1\ge6$. If $|\Delta|\ge4$, we take them in any coordinates other than $j$, where $g$ has mass $r-|x|\ge7$. Here $k=2$. Write $a_s=a-s\sigma e_j$. The steps of $a$ move its $j$-th coordinate away from $x$, which is the $j$-th coordinate of every $g_t$, so $|a_s-g_t|=|a-g_t|+s$. If $|\Delta|=2$, the steps of $g$ are in coordinates where $a$ and $g$ agree and $a$ never moves, so $|a-g_t|=2+t$. If $|\Delta|\ge4$, then $|a-g_t|\ge|\Delta|-t\ge2$. This proves (E2). Since $|x|\ge2$, the sites $a_1,g_1$ lie in layer $r-1$ and $a_2,g_2$ in layer $r-2$, which gives (E1). For (E3), the $j$-th coordinates of $a_2$ and $g_2$ are $x-2\sigma$ and $x$, and $s=x-\sigma$ satisfies $|s|=|x|-1\le r-8\le(r-2)-5$. \textbf{Type I: some coordinate $p$ has $\max(|a_p|,|g_p|)\ge r-6$.} Then $a_p$ and $g_p$ have a common sign $\sigma_p$ and $|a_p|,|g_p|\ge r-7$: this follows from Step 0 if $|\Delta_p|\ge2$, and directly if $|\Delta_p|\le1$. Write $\operatorname{down}x=x-\sigma_pe_p$. Each down step below is applied to a site with $|x_p|\ge(r-7)-3\ge10$, so it is an inward step. The coordinates $j\ne p$ are called \textit{small}. The small parts of $a$ and $g$ have norm at most $70$, and in particular $\mu(V\ne\emptyset)>0$. \textit{Conditioning.} Let $\nu=\mu(\,\cdot\mid V\ne\emptyset)$. The event $\{V\ne\emptyset\}$ is invariant, so $\nu$ is invariant and $\nu(\mathcal X_{\rho_0})=1$. On the event $\{0<|V|<\infty\}$ let $\ell$ be the lexicographically least site of $V$. Then $\ell\circ T_z=\ell+z$, so by invariance $\nu(\ell=z)=\nu(\ell=0)$ for all $z$. These numbers sum to $\nu(0<|V|<\infty)\le1$, so they all vanish, and $V$ is $\nu$-almost surely infinite.\qedsq \subsection*{4.3. Mass transport} A \textit{covariant partial map} assigns to each $\xi\in\Xi$ a map $\varphi_\xi$ from a set $\operatorname{dom}\varphi_\xi\subseteq\mathbb Z^d$ to $\mathbb Z^d$, measurably in $\xi$, such that $\operatorname{dom}\varphi_{T_z\xi}=\operatorname{dom}\varphi_\xi+z$ and $\varphi_{T_z\xi}(x+z)=\varphi_\xi(x)+z$ for all $z$ and $x$. \textbf{Lemma 4.2} (mass transport on $\mathbb Z^d$). \textit{Let $\nu$ be a translation-invariant probability measure on $\Xi$.} \begin{enumerate} \item{} \textit{If $F:\mathbb Z^d\times\mathbb Z^d\times\Xi\to[0,\infty]$ is measurable and $F(x+z,y+z,T_z\xi)=F(x,y,\xi)$ for all $x,y,z$ and $\xi$, then $\mathbb E_\nu\sum_{y}F(0,y,\xi)=\mathbb E_\nu\sum_{x}F(x,0,\xi)$.} \item{} \textit{If $\varphi$ is a covariant partial map and $\varphi_\xi$ is injective for $\nu$-almost every $\xi$, then $\nu\bigl(0\in\varphi_\xi(\operatorname{dom}\varphi_\xi)\bigr)=\nu(0\in\operatorname{dom}\varphi_\xi)$.} \end{enumerate} \textit{Proof.} (1) For each $y$, covariance with $z=-y$ and invariance give $\mathbb E\,F(0,y,\xi)=\mathbb E\,F(-y,0,T_{-y}\xi)=\mathbb E\,F(-y,0,\xi)$. Sum over $y$. (2) Apply (1) to $F(x,y,\xi)=\mathbf 1[x\in\operatorname{dom}\varphi_\xi,\ \varphi_\xi(x)=y]$. The left side is $\nu(0\in\operatorname{dom}\varphi_\xi)$. The right side is the expected number of preimages of $0$, which is almost surely at most one and equals the indicator of $0\in\varphi_\xi(\operatorname{dom}\varphi_\xi)$.\qedsq Part (1) is the mass-transport principle of Benjamini, Lyons, Peres and Schramm [BLPS] (see also [LP, Chapter 8]) in its simplest case, the group $\mathbb Z^d$. \subsection*{4.4. Shrinking towards the apex} The next lemma is deterministic and holds for every pair $(V,\prec)$. For $R\ge2$ let $\mathcal D_R$ be the set of sites $x$ such that $B_R(x)$ is corner-paired. For $R\ge r\ge2$ and $x\in \mathcal D_R$ put \[ \Psi_{R\to r}(x)=x+(R-r)\,\sigma_R(x). \] Since $\mathcal D_R$ and $\sigma_R(x)$ are determined by the restriction of $(V,\prec)$ to $B_R(x)$, the map $\Psi_{R\to r}$ is a covariant partial map with domain $\mathcal D_R$. \textbf{Lemma 4.3.} \textit{Let $R\ge r\ge2$ and $x\in \mathcal D_R$, with pole $\sigma$, apex $C=x+R\sigma$ and partner $N=x+(R-1)\sigma+\delta$. Put $y=\Psi_{R\to r}(x)$.} \begin{enumerate} \item{} \textit{$B_r(y)\subseteq B_R(x)$, and $B_r(y)$ contains $C$ and $N$. Consequently $B_r(y)$ has the same first and last sites and the same window as $B_R(x)$, and it is corner-paired with pole $\sigma$ and apex $C$.} \item{} \textit{$\Psi_{R\to r}$ is injective on $\mathcal D_R$.} \end{enumerate} \textit{Proof.} (1) If $|z-y|\le r$ then $|z-x|\le r+(R-r)=R$. The vectors $C-y=r\sigma$ and $N-y=(r-1)\sigma+\delta$ have norm $r$. The first site of $V\cap B_R(x)$ belongs to the smaller set $V\cap B_r(y)$, so it is also the first site of $V\cap B_r(y)$, and the same holds for the last site. Hence the windows coincide, and $\{C,N\}=\{y+r\sigma,\ y+(r-1)\sigma+\delta\}$ is a corner pair of $S_r(y)$ with pole $\sigma$. (2) If $\Psi_{R\to r}(x)=y$, then by (1) the ball $B_r(y)$ is corner-paired with pole $\sigma_R(x)$. As $r\ge2$, the pole of $B_r(y)$ is unique, so $\sigma_R(x)=\sigma_r(y)$ and $x=y-(R-r)\sigma_r(y)$ is determined by $y$.\qedsq For $r=1$ the pole is not unique and injectivity can fail, which is why all radii below are at least $2$. \textbf{Proposition 4.4.} \textit{Let $\nu$ be a translation-invariant probability measure on $\Xi$ with $\nu(\mathcal X_{\rho_0})=1$ under which $V$ is almost surely infinite. Then $\nu$-almost surely, for every $y\in\mathbb Z^d$ and every $r\ge2$, the ball $B_r(y)$ is corner-paired, and for every $R\ge r$ the ball $B_R(y-(R-r)\sigma_r(y))$ is corner-paired with the same window and the same apex $y+r\sigma_r(y)$ as $B_r(y)$.} \textit{Proof.} Fix $r\ge2$, and for $R\ge r$ let $\mathcal E_R$ be the event $\{0\in\Psi_{R\to r}(\mathcal D_R)\}$. First let $R\ge\max(r,\rho_0)$, and let $A_R$ be the set of sites $x$ with $|V\cap B_R(x)|\ge3$. By (Cor), $A_R\subseteq \mathcal D_R$ almost surely. Since $V$ is almost surely infinite, $\beta_R:=\nu(0\in A_R)$ tends to $1$ as $R\to\infty$. By Lemma 4.3(2) and Lemma 4.2(2), $\nu(\mathcal E_R)=\nu(0\in \mathcal D_R)\ge\beta_R$. Next, $\mathcal E_{R'}\subseteq \mathcal E_R$ whenever $R'\ge R\ge r$. Indeed, let $0=\Psi_{R'\to r}(x)$ with $x\in \mathcal D_{R'}$ and pole $\sigma$. By Lemma 4.3(1), the site $x'=\Psi_{R'\to R}(x)=x+(R'-R)\sigma$ lies in $\mathcal D_R$ with pole $\sigma$, and $\Psi_{R\to r}(x')=x+(R'-r)\sigma=0$. Hence $\nu(\mathcal E_R)\ge\nu(\mathcal E_{R'})\ge\beta_{R'}$ for every $R'\ge\max(R,\rho_0)$, so $\nu(\mathcal E_R)=1$ for every $R\ge r$, and the events $\mathcal E_R$, $R\ge r$, hold simultaneously almost surely. On this event, fix $R\ge r$ and $x\in \mathcal D_R$ with $\Psi_{R\to r}(x)=0$. By Lemma 4.3(1), $B_r(0)$ is corner-paired with pole $\sigma_R(x)$, and $\sigma_R(x)=\sigma_r(0)$ by uniqueness of the pole. Thus $x=-(R-r)\sigma_r(0)$, and $B_R(x)$ has the same window as $B_r(0)$ and the apex $x+R\sigma_r(0)=r\sigma_r(0)$. This proves the statement for $y=0$ and the fixed $r$. By translation invariance it holds for each $y$ almost surely, and a countable intersection over $y$ and $r$ completes the proof.\qedsq \subsection*{4.5. One pole per apex} \textbf{Lemma 4.5.} \textit{Let $(V,\prec)$ be any pair, and let $B_1=B_{r_1}(c_1)$ and $B_2=B_{r_2}(c_2)$ be corner-paired balls with $r_1,r_2\ge2$, poles $\pi_1\ne\pi_2$, apexes $C_1,C_2$ and partners $N_1=C_1-\pi_1+\delta_1$, $N_2=C_2-\pi_2+\delta_2$. Then neither of the following holds:} \begin{itemize} \item{} \textit{(perpendicular) $\pi_1\perp\pi_2$ and $C_2=C_1+t(\pi_2-\pi_1)$ for an integer $t$ with $2\le t\le\min(r_1,r_2)$;} \item{} \textit{(antipodal) $\pi_2=-\pi_1=:-\sigma$ and $C_1=C_2+t\sigma$ for an integer $t$ with $2\le t\le\min(r_1,r_2)+1$.} \end{itemize} \textit{Proof.} We first show that in either situation each ball contains both ports of the other. In the perpendicular situation $c_1=C_1-r_1\pi_1$ and $c_2=C_2-r_2\pi_2$, so \[ C_2-c_1=(r_1-t)\pi_1+t\pi_2,\qquad N_2-c_1=(r_1-t)\pi_1+(t-1)\pi_2+\delta_2, \] \[ C_1-c_2=t\pi_1+(r_2-t)\pi_2,\qquad N_1-c_2=(t-1)\pi_1+(r_2-t)\pi_2+\delta_1 . \] The first and third vectors have norms $r_1$ and $r_2$. Since $\delta_2\perp\pi_2$, either $\delta_2=-\pi_1$, or $\delta_2=\pi_1$, or $\delta_2$ is orthogonal to both poles. The norm of $N_2-c_1$ is then $|r_1-t-1|+t-1$, $r_1$ or $r_1$, and in all cases at most $r_1$. The fourth vector is treated in the same way and has norm at most $r_2$. In the antipodal situation $c_1=C_1-r_1\sigma$, $c_2=C_2+r_2\sigma$, $N_1=C_1-\sigma+\delta_1$ and $N_2=C_2+\sigma+\delta_2$, so \[ \resizebox{473.2pt}{!}{$\displaystyle C_2-c_1=(r_1-t)\sigma,\quad N_2-c_1=(r_1-t+1)\sigma+\delta_2,\quad C_1-c_2=(t-r_2)\sigma,\quad N_1-c_2=(t-1-r_2)\sigma+\delta_1 .$} \] Since $2\le t\le\min(r_1,r_2)+1$, their norms $|r_1-t|$, $r_1-t+2$, $|t-r_2|$ and $r_2-t+2$ are at most $r_1$, $r_1$, $r_2$ and $r_2$. So $C_2,N_2\in B_1$ and $C_1,N_1\in B_2$. The ports of $B_2$ are sites of $V\cap B_1\subseteq W(B_1)$, and $W(B_1)$ is an interval of $(V,\prec)$, so $W(B_2)\subseteq W(B_1)$. By symmetry $W(B_1)=W(B_2)$, so the two balls have the same first and last sites, and $\{C_1,N_1\}=\{C_2,N_2\}$. Since $t\ne0$ we have $C_1\ne C_2$, hence $C_1=N_2$ and $C_2=N_1$. In the perpendicular situation, $C_2-C_1=N_1-C_1=-\pi_1+\delta_1$ has $\pi_1$-coefficient $-1$, while $t(\pi_2-\pi_1)$ has $\pi_1$-coefficient $-t$, so $t=1$, a contradiction. In the antipodal situation, $C_1-C_2=N_2-C_2=\sigma+\delta_2$ would equal $t\sigma$, which is impossible because $\delta_2\perp\sigma$.\qedsq The hypothesis $t\ge2$ cannot be dropped in the perpendicular situation. For $t=1$ the two balls can have the same port pair $\{C_1,C_2\}$. For $t=0$ a single site can be the apex for both poles, with the common port pair $\{C,C-\pi_1-\pi_2\}$. So Lemma 4.5 alone does not exclude two poles at one apex, and the next lemma uses stationarity. \textbf{Lemma 4.6.} \textit{Let $\nu$ be as in Proposition 4.4. Then $\nu$-almost surely no site is the apex of two corner-paired balls, of radii at least $2$, with different poles.} \textit{Proof.} We work on the full-measure event of Proposition 4.4. Let $C$ be the apex of a corner-paired ball $B_r(y)$ with pole $\pi$ and $r\ge2$. Then for every $r'\ge2$ the ball $B_{r'}(C-r'\pi)$ is corner-paired with pole $\pi$ and apex $C$: for $r'\ge r$ this is Proposition 4.4, since $y-(r'-r)\pi=C-r'\pi$, and for $2\le r'0$ for all $p\in(\pi(s),p_c(G))$. Since $E(\omega,\alpha)$ is increasing in $\alpha$, $\theta_E(p,\cdot)$ is non-decreasing, so the case $s=1$ follows from $s=\frac12$. \textbf{Remark 6.3.} The definition of $p_c(G)$ does not depend on the root: by finite energy, $\theta_x(p)>0$ for one site $x$ if and only if it holds for every site. For a single non-monotone enhancement the positivity of $\theta_E$ can depend on the root on general periodic lattices, so Conjecture 1 is stated at the root $o$. \textbf{Remark 6.4} (all roots). Let $\phi$ be an automorphism of $G$ with $\phi\circ\tau_y=\tau_{Ay}\circ\phi$ for some $A\in GL(n,\mathbb Z)$ and all $y$, and put $E^\phi_v(\omega)=\phi\bigl(E_{\phi^{-1}v}(\phi^{-1}\omega)\bigr)$. Then $E^\phi$ is an enhancement with range $r_E$, it is essential if and only if $E$ is, and $E^\phi(\phi\omega,\phi\alpha)=\phi E(\omega,\alpha)$, so the probability that $\phi(x)$ lies in an infinite component of $E^\phi(\omega,\alpha)$ equals that for $x$ and $E(\omega,\alpha)$. If such automorphisms act transitively on $V$, then Conjecture 1 for all essential enhancements at the root $o$ gives the same conclusion at every root. This applies to TRI (translations), to HEX ($\phi(z,i)=(-z,1-i)$ with the translations) and to KAG (the automorphism $a(z)\mapsto b(Mz)$, $b(z)\mapsto c(Mz)$, $c(z)\mapsto a(Mz)$, where $Me_1=-e_1+e_2$ and $Me_2=-e_1$, with the translations). The verification is routine and is given in Appendix B.7. The same applies to the lattices of Section 12: on those with $k=1$ the translations act transitively, and on the others the translations together with one further automorphism $\phi$ with $\phi\circ\tau_y=\tau_{Ay}\circ\phi$ act transitively (Sections 12.4–12.7). \subsection*{6.4. The probabilistic step} \textbf{Theorem 6.5} (the probabilistic step; [AG, §3], in the form of [BBR, pp. 4–6]). \textit{Let $G$ be a periodic lattice with $p_c(G)<1$, and let $E$ be an enhancement of site percolation on $G$. If Claim 5 holds for $E$, then Conjecture 1 holds for $E$.} The proof follows the four steps recalled in Section 2.2, with $N_p(L)$ and $N_s(L)$ as there. Claim 5 implies Claim 4 because the enhanced configuration is increasing in $\alpha$ [BBR, pp. 5–6]; Claim 4 and a finite-to-one modification map give a continuous positive function $g$ on $(0,1)^2$ with $\mathbb E_{p,s}N_s(L)\ge g(p,s)\,\mathbb E_{p,s}N_p(L)$ for $L\ge L_0$ [BBR, (2)]; a Margulis–Russo formula for the polynomial $\tau_L(p,s)=\mathbb P_{p,s}(R_L(E(\omega,\alpha)))$ turns this into $\partial_s\tau_L\ge g\,\partial_p\tau_L$ [BBR, (3)]; and comparing $\tau_L$ along curves independent of $L$, as $L\to\infty$, gives the conclusion [AG, pp. 822–823]. On a periodic lattice the only new points are a bound $M_R=\max_c|B_R(c)|\le(\Delta+1)^R$ on the size of balls that is uniform over all sites, which makes $g$ independent of $L$, and the locality of enhancements and crossing events for graph balls. The differential inequality on graphs of bounded degree is also proved, for the enhancement used there, in [MS, §6]. A complete proof is given in Appendix B, with the explicit value \[ \resizebox{473.2pt}{!}{$\displaystyle \pi(s)=p_c-\min\Bigl(\frac{p_c}{2},\ \frac{g_0s_0}{2},\ \frac{1-p_c}{2}\Bigr),\qquad s_0=\min\bigl(s,\tfrac12\bigr),\qquad g_0=\frac{1}{M_R4^{M_R}}\Bigl(\frac{\min(p_c,1-p_c)\,s_0}{4}\Bigr)^{M_R},$} \] where $p_c=p_c(G)$ and $R$ is the radius in Claim 5. \subsection*{6.5. Critical points} \textbf{Proposition 6.6.} \textit{Let $G$ be one of the lattices of Theorem 1.2. Then $0p_c^{\mathrm{site}}(\mathbb Z^2)^{1/b}$, then $\theta_{\mathbb Z^2}(q^b)>0$, hence $\theta_G(q)>0$.\qedsq \textit{Proof of Proposition 6.6.} The lower bounds are Lemma 6.7. For the upper bounds we apply Lemma 6.8 with the following branch sets. For TRI take $\beta(z)=\{(z,0)\}$ (adjacent to $\beta(z+e_j)$ since $e_j$ is a lattice step), for HEX $\beta(z)=\{A(z),B(z)\}$ (connected as $A(z)\sim B(z)$; $B(z)\sim A(z+e_j)$), and for KAG $\beta(z)=\{a(z),b(z),c(z)\}$ (an up triangle; $b(z)\sim a(z+e_1)$ and $c(z)\sim a(z+e_2)$ by the down triangles at $z+e_1$ and $z+e_2$). For KING take $\beta(z)=\{(z,0)\}$ ($\mathbb Z^2$ is a subgraph), and for FCC $\beta(z)=\{((z_1,z_2,0),0)\}$ ($\pm e_1,\pm e_2$ are lattice steps). For ELTRI, SNUBSQ, SNUBHEX and RTH take for $\beta(z)$ the $k$ sites of the cell $z$, $k=2,4,6,6$. By the neighbour tables of Sections 12.4–12.7 they span a connected subgraph: an edge for ELTRI, a $4$-cycle $(z,0),(z,1),(z,2),(z,3)$ for SNUBSQ and a $6$-cycle $(z,0),\dots,(z,5)$ for SNUBHEX and RTH. Some site of $\beta(z)$ is adjacent to some site of $\beta(z+e_j)$: for ELTRI $((z),0)\sim((z+e_1),0)$ and $((z),1)\sim((z+e_2),0)$; for SNUBSQ $(z,0)\sim(z+e_1,1)$ and $(z,3)\sim(z+e_2,0)$; for SNUBHEX $(z,0)\sim(z+e_1,2)$ and $(z,0)\sim(z+e_2,4)$; for RTH $(z,0)\sim(z+e_1,2)$ and $(z,0)\sim(z+e_2,4)$. Lemma 6.8 applies with $b=|\beta(z)|$.\qedsq The bounds are far from the true values, but only the strict inequalities matter here. \section*{7. The reduction on periodic lattices} Throughout this section $G$ is a periodic lattice and $E$ is an essential enhancement of site percolation on $G$ with range $r_E$ and essential orbit $i$. Red and green sets are denoted $\mathsf R$ and $\mathsf G$, to avoid a clash with the lattice $G$ and with the radius $R$ of Claim 5. We prove Theorem A below by going through the proof of Theorem 2.13 (Sections 2.3–2.7) and listing what changes; the steps that change in substance (the gadget, the ports, the main case and the two boundary cases) are proved in full. \subsection*{7.1. Elementary facts} The following facts hold on every periodic lattice. They are the only properties of balls used in this section. \begin{itemize} \item{} \textbf{(F1)} \textit{Balls are finite and connected, $B_{r+1}(c)=N[B_r(c)]$, and $S_r(c)\ne\emptyset$ for all $c$ and $r$.} A site at distance $r+1$ from $c$ is adjacent to the second-to-last site of a geodesic, which lies at distance $r$; finiteness follows by induction from local finiteness. A geodesic from $c$ to a site of $B_r(c)$ stays in $B_r(c)$. Since $V$ is infinite, some site has distance at least $r$ from $c$, and a geodesic to it meets $S_r(c)$. \item{} \textbf{(F2)} \textit{If $x\sim y$, then $|d(x,c)-d(y,c)|\le1$ for every site $c$.} \item{} \textbf{(F3)} \textit{$N[B_\rho(c)]\subseteq B_{\rho+1}(c)$ and $B_{\rho-1}(c)\subseteq\operatorname{int}B_\rho(c)$ for $\rho\ge1$.} \item{} \textbf{(F4)} \textit{Each $\tau_z$ is an automorphism of $G$.} Hence it preserves distances, balls, spheres, paths, induced paths, connected sets, pockets and escapable pairs (Definition 7.1). \item{} \textbf{(F5)} \textit{If $d(x,c)=r\ge1$, then $x$ has a neighbour in $S_{r-1}(c)$.} \item{} \textbf{(F6)} (discrete intermediate values) \textit{Let $(y_0,\dots,y_q)$ be a walk and $s$ an integer with $d(y_0,c)\le s\le d(y_q,c)$. Then the least $j$ with $d(y_j,c)\ge s$ satisfies $d(y_j,c)=s$.} In particular, for $0\le n\le L$ every path from $o$ to a site of $S_L$ meets $S_n$. \item{} \textbf{(F7)} \textit{Every walk from $x$ to $y$ contains a path from $x$ to $y$ all of whose sites are sites of the walk.} \item{} \textbf{(F8)} (locality) \textit{$E_v(\omega)\subseteq B_{r_E}(v)$, and $E_v(\omega)=E_v(\tilde\omega)$ whenever $\omega\cap B_{r_E}(v)=\tilde\omega\cap B_{r_E}(v)$.} This follows from Definition 6.2 and (F4). \item{} \textbf{(F9)} \textit{For every site $c$ and every orbit $j$ there is a site $c'$ of the orbit $j$ with $d(c,c')\le\kappa(G)$.} \item{} \textbf{(F10)} \textit{$R_L$ is increasing, and $\alpha\mapsto E(\omega,\alpha)$ is increasing.} (The map $\omega\mapsto E(\omega,\alpha)$ need not be.) \end{itemize} \subsection*{7.2. The lattice hypotheses and Theorem A} \textbf{Definition 7.1} (escapable pair). Let $c'\in V$ and $m\ge1$. An ordered pair $(t_+,t_-)$ of distinct sites of $S_m(c')$ is \textit{escapable} if there are $u_+,u_-\in S_{m+1}(c')$ with $u_+\sim t_+$ and $u_-\sim t_-$ such that $\{t_+,u_+\}$ and $\{t_-,u_-\}$ do not touch. \textbf{Definition 7.2} (design). Let $c,c'\in V$ and $\rho,m\in\mathbb N$ with $B_m(c')\subseteq B_{\rho-1}(c)$, let $a\ne g$ be sites of $S_\rho(c)$, and let $(t_+,t_-)$ be a pair of sites of $S_m(c')$. A \textit{design} for $(a,g;t_+,t_-)$ in $B_\rho(c)$ is a pair of sets $\mathsf R,\mathsf G\subseteq B_\rho(c)$ such that: \begin{itemize} \item{} (D1) $a\in\mathsf R$, $g\in\mathsf G$, and $\mathsf R$ and $\mathsf G$ are connected; \item{} (D2) $\mathsf R\setminus\{a\}\subseteq B_{\rho-1}(c)$ and $\mathsf G\setminus\{g\}\subseteq B_{\rho-1}(c)$; \item{} (D3) $\mathsf R\cap B_m(c')=\{t_+\}$ and $\mathsf G\cap B_m(c')=\{t_-\}$, or $\mathsf R\cap B_m(c')=\{t_-\}$ and $\mathsf G\cap B_m(c')=\{t_+\}$; \item{} (D4) $\mathsf R$ and $\mathsf G$ do not touch. \end{itemize} An \textit{exceptional family} is a family $X=(X_\rho(c))_{c\in V,\rho\ge2}$, where $X_\rho(c)$ is a set of unordered pairs of distinct sites of $S_\rho(c)$. We consider the following properties of $G$. All constants are integers. \begin{itemize} \item{} \textbf{(B1)} (gadget radii) There is an infinite set $M(G)$ of radii such that for every $m\in M(G)$ and every orbit representative $x_j$, the sphere $S_m(x_j)$ has no pockets and every pair of distinct non-adjacent sites of $S_m(x_j)$ is escapable. \item{} \textbf{(B2)} (inner boundary) There are $n_0,t_0,c_2\ge1$ and an infinite set $M'(G)\subseteq\mathbb N$ whose consecutive elements differ by at most $c_2$, such that for every $n\in M'(G)$ with $n\ge n_0$: $S_n$ has no pockets, the annulus $S_{n+1}\cup\dots\cup S_{n+t_0}$ is connected, and every site $v$ with $|v|\le n$ satisfies $d(v,S_n)\le n-|v|+c_2$. \item{} \textbf{(B3)} (outer boundary) There are $0\le c_3r_E$, and $\omega_g\cap S_m(x_i)=\{t_+,t_-\}$ with $(t_+,t_-)$ an escapable pair of $S_m(x_i)$;} \item{} \textit{$\omega_g$ contains no path from $t_+$ to $t_-$;} \item{} \textit{$\omega_g\cup E_i(\omega_g)$ contains a path from $t_+$ to $t_-$;} \item{} \textit{$E_i(\omega_g)\subseteq B_{r_E}(x_i)\subseteq B_{m-1}(x_i)$.} \end{enumerate} \textit{Moreover $t_+$ and $t_-$ are distinct and non-adjacent. Without (B1) the same construction works for every integer $m\ge m_E$, with “escapable” in 1 replaced by “distinct and non-adjacent”.} \textit{Proof.} \textit{Step 1: a thin essential configuration.} Since $i$ is an essential orbit, there are a configuration $\omega^0$ that contains no doubly infinite path and a doubly infinite path $\Gamma:\mathbb Z\to V$ in $\omega^0\cup E_i(\omega^0)$. Put $X_0=E_i(\omega^0)$ and \[ \omega_1=\bigl(\omega^0\cap B_{r_E}(x_i)\bigr)\cup\bigl(\Gamma(\mathbb Z)\setminus B_{r_E}(x_i)\bigr). \] (a) $\omega_1\subseteq\omega^0$, because $X_0\subseteq B_{r_E}(x_i)$ gives $\Gamma(\mathbb Z)\setminus B_{r_E}(x_i)\subseteq\omega^0$. So $\omega_1$ contains no doubly infinite path. (b) $\omega_1\cap B_{r_E}(x_i)=\omega^0\cap B_{r_E}(x_i)$, so $E_i(\omega_1)=X_0$. (c) $\Gamma(\mathbb Z)\subseteq\omega_1\cup X_0$. \textit{Step 2: the two tails.} Let $J=\{j:\Gamma(j)\notin\omega_1\}$. By (c), $\Gamma(J)\subseteq X_0$, which is finite, so $J$ is finite, and $J\ne\emptyset$ by (a). Let $j_-=\min J$, $j_+=\max J$, $T_+=\{\Gamma(j):j>j_+\}$ and $T_-=\{\Gamma(j):ji_1$ with $q_i\in T_-$, and write $q_{i_1}=\Gamma(p)$, $q_{i_2}=\Gamma(p')$. The sites $q_i$ with $i_1j_++1$, $\Gamma(p_+-1)\in B_{m-1}(x_i)$ and $t_+\in S_m(x_i)$; all $\Gamma(j)$ with $j_+r_E$, and $\omega_g\cap S_m(x_i)=\{t_+,t_-\}$. The sites $t_\pm$ are distinct, and non-adjacent by the claim. By (B1), $(t_+,t_-)$ is escapable if $m\in M(G)$. (2) $\omega_g\subseteq\omega_1$, so a path in $\omega_g$ from $t_+$ to $t_-$ would contradict the claim. (4) As $r_E\le m-1$ and $t_\pm\notin B_{m-1}(x_i)$, we have $\omega_g\cap B_{r_E}(x_i)=\omega_1\cap B_{r_E}(x_i)$, so $E_i(\omega_g)=X_0\subseteq B_{r_E}(x_i)$ by (b). (3) The sites of $(\Gamma(p_-),\dots,\Gamma(p_+))$ lie in $\omega_g\cup X_0$: the ends are $t_\pm$; for $p_-r_E$ for every $a\in\alpha$. \end{itemize} If $K<|v|\min M'(G)$ there is $n\in M'(G)$ with $N\le n\le N+c_2-1$. (c) If (B3) holds with $(c_3,c_4,K_3)$, then for every integer $j\ge0$ it holds with $(c_3+j,c_4+j,K_3)$.} \textit{Proof.} (a) If the ends of $P$ lie outside $B_K(v)$, they lie outside $B_{K_*}(v)$, and the ball found for $K_*$ lies in $B_{K_*}(v)\subseteq B_K(v)$. (b) Let $n$ be the least element of $M'(G)$ with $n\ge N$. It has a predecessor $n^-\min M'(G)$, Lemma 7.11(b) gives $n\in M'(G)$ with $K+11\le n\le K+10+c_2$, and $n\ge n_0$. By (B2), $S_n$ has no pockets, $A=S_{n+1}\cup\dots\cup S_{n+t_0}$ is connected, and there is $w\in S_n$ with \[ d(v,w)\le n-|v|+c_2\le K+10+2c_2. \tag{7.3} \] Put $\Lambda=B_{n+t_0}$. For $a\in\alpha$ and $x\in\Lambda$ we have $d(a,x)\ge|a|-|x|>(R-K-10)-(n+t_0)\ge K>r_E$, so Lemma 7.3 gives $Y'$ with $Y'\cap\Lambda=\emptyset$ and $E(\hat\omega,\alpha)=\hat\omega\cup Y'$ whenever $\hat\omega$ agrees with $\omega$ outside $\Lambda$. Let \[ \tilde\omega=(\omega\setminus\Lambda)\cup(\Lambda\setminus S_n)\cup\{w\}. \] \textit{$E(\tilde\omega,\alpha)$ satisfies $R_L$.} Let $\Gamma$ be a path from $o$ to $y\in S_L$ in $E(\omega\cup\{v\},\alpha)\subseteq\omega\cup\{v\}\cup Y'$. As $L>n+t_0$, the last site $x'$ of $\Gamma$ in $\Lambda$ lies in $S_{n+t_0}\subseteq A$, and the part after it lies in $(\omega\setminus\Lambda)\cup Y'$. Inside $\Lambda$, the connected set $B_{n-1}\ni o$, a neighbour $w^-\in S_{n-1}$ of $w$ (F5), the site $w$, a neighbour $w^+\in S_{n+1}$ of $w$ (no pockets) and the connected set $A\ni w^+,x'$ are open in $\tilde\omega$. Concatenation and (F7) give the claim. \textit{$E(\tilde\omega\setminus\{w\},\alpha)$ does not satisfy $R_L$}, because it contains no site of $S_n$, and every path from $o$ to $S_L$ meets $S_n$ (F6). So $w$ is p-pivotal in $(\tilde\omega,\alpha)$. Now $K<|w|=n\le K+10+c_2R-(K+c_4^*-1)-(K+c_4^*)>r_E$, so Lemma 7.3 gives $Y$ with $Y\cap\Omega'=\emptyset$ and $E(\hat\omega,\alpha)=\hat\omega\cup Y$ whenever $\hat\omega$ agrees with $\omega$ outside $\Omega'$. \textit{Step 3.} Let $u=q_k$ be the first site of $Q$ in $\Omega'$. By (F2), $d(u,w)=K+c_4^*$, so $u\ne v$, $u$ comes before $v$ on $Q$, $|u|0$ such that for all $\rho\ge\rho_X$ and all $c_1,c_2$ for which $A_j:=A_\rho(c_j,\pi_j)$ is defined ($j=1,2$) and $A_2=\tau_{tu}A_1$ for an integer $t$ with $t_0\le t\le\lambda\rho$, we have $d(A_1,A_2)>p_0$, $B_\rho(c_1)\supseteq\{A_2\}\cup N_\rho(c_2,\pi_2)$ and $B_\rho(c_2)\supseteq\{A_1\}\cup N_\rho(c_1,\pi_1)$. \end{itemize} For every ordered pair $(\pi_1,\pi_2)$ of distinct labels we fix one triple $(u,t_0,\lambda)=(u(\pi_1,\pi_2),t_0(\pi_1,\pi_2),\lambda(\pi_1,\pi_2))$ as in (X4), its \textit{data}. The site $A_\rho(c,\pi)$ is an \textit{apex} of $B_\rho(c)$, and the sites of $N_\rho(c,\pi)$ are its \textit{partners}. On $\mathbb Z^d$ the corner pairs of Definition 2.11 satisfy (X1)–(X4) with apexes $c+\rho\pi$ (Lemma 8.14). \textbf{Theorem C.} \textit{Let $G$ be a periodic lattice with $\delta(G)\ge3$, and let $X$ be an exceptional family satisfying (X1)–(X4) with constants $\rho_X,p_0$. Then CL$_X$ holds. More precisely, for every integer $\rho_0\ge\rho_X$ the conclusion of CL$_X$ holds with the constant $K=K(\rho_0)$ of (8.1) below. The covariance (X1) and the covariance clause of (X2) are not used.} The conclusion for $\rho_0\ge\rho_X$ implies it for every $2\le\rho_0<\rho_X$, since the ball found has radius at least $\rho_X>\rho_0$; so CL$_X$ holds for every $\rho_0\ge2$. If the conclusion holds with some $K$, it holds with every larger $K$. \textit{Idea of the proof.} Suppose that the conclusion fails for a path $P$ through $v$. Then every ball of radius at least $\rho_0$ well inside $B_K(v)$ that contains at least three sites of $P$ has an exceptional port pair, one of whose sites is an apex, and the centre is determined by the apex and its label. Fix a large radius $r$ and a box around $v$ that is much larger than $r$, and compare three counts. \textit{Density:} the balls of a radius proportional to the box, centred in the box, all contain $v$ and its two neighbours on $P$; by the injectivity in (X2), $P$ occupies a fixed proportion of a larger box, independent of $r$. \textit{Degree:} every site of $P$ has a neighbour off $P$, and this neighbour is the centre of a ball of radius $r$ with at least three sites of $P$; so the set $\mathcal W_r$ of such centres exceeds $P$ by a fixed proportion. \textit{Apexes:} the centres in $\mathcal W_r$ map injectively to pairs (apex, label), and apexes are sites of $P$; a site that is an apex for two labels is rare, because by (X3) and (X4) two such sites cannot lie between $t_0$ and $\lambda r$ steps apart along a line in direction $u$. So $|\mathcal W_r|$ exceeds $|P|$ in the box only by a boundary term and a term of relative order $1/r$. This contradicts the degree count when $r$ is large and the box is large compared with $r$. The argument is a finitary counting, like the formal proof of the Corner Lemma (Section 14), and it replaces the stationary limit and the mass transport of Sections 4.2–4.6. It needs neither the shrinking map of Lemma 4.3 nor the uniqueness of the pole at an apex (Lemma 4.6). \subsection*{8.2. Boxes and two lattice constants} For $q\in\mathbb Z^n$ and an integer $h\ge0$ let $Q_h(q)=\{x\in V:\|\bar x-q\|_\infty\le h\}$, so that $|Q_h(q)|=k(2h+1)^n$. Put \[ \ell=\max\{\|\bar x-\bar y\|_\infty:x\sim y\},\qquad\gamma=\max\{d(x,y):\|\bar x-\bar y\|_\infty\le1\}. \] \textbf{Lemma 8.1.} \textit{(1) $\ell$ and $\gamma$ are positive integers. (2) $\|\bar x-\bar y\|_\infty\le\ell\,d(x,y)$. (3) $d(x,y)\le\gamma\max\{1,\|\bar x-\bar y\|_\infty\}$. (4) If $x\in Q_h(q)$ and $d(x,y)\le s$, then $y\in Q_{h+\ell s}(q)$. If $h\ge1$, then $Q_h(\bar v)\subseteq B_{\gamma h}(v)$ for every site $v$.} \textit{Proof.} (1) By translation invariance the maxima are over finitely many pairs, involving the sites $(0,i)$ and their neighbours, resp. the sites $(q,j)$ with $q\in\{-1,0,1\}^n$; they are finite by local finiteness and connectedness. If $\ell=0$, the finite set $\{0\}\times\{0,\dots,k-1\}$ would be a union of components of $G$. Also $\gamma\ge d((0,0),(e_1,0))\ge1$. (2) Sum along a geodesic. (3) If $m=\|\bar x-\bar y\|_\infty\ge2$, move from $\bar x$ to $\bar y$ through cells $q_0,\dots,q_m$ with $\|q_j-q_{j-1}\|_\infty\le1$ (changing every coordinate that is not yet correct by one unit at each step), and use sites $(q_j,0)$ for $0p_0\ge0$, so $A_1=N_2$, and then $d(A_1,A_2)=d(N_2,A_2)\le p_0$, a contradiction.\qedsq The same argument applies to two labellings of one ball, so exceptional pairs whose two sites are both apexes need no separate treatment. \subsection*{8.4. Consequences of a failure} In this subsection $\rho_0\ge\rho_X$ and $K\ge0$ are integers, $P=(x_0,\dots,x_N)$ is an induced path, $v$ is a site of $P$ with $x_0,x_N\notin B_K(v)$, and we assume that the conclusion of CL$_X$ fails: \begin{quote} \textbf{(Cor)} for every $c\in V$ and every $\rho\ge\rho_0$ with $B_{\rho+1}(c)\subseteq B_K(v)$ and $|P\cap B_\rho(c)|\ge3$, we have $\{a(c,\rho),g(c,\rho)\}\in X_\rho(c)$. \end{quote} We write $Q_h=Q_h(\bar v)$. For $r\ge1$ let $\mathcal W_r=\{y\in V:|P\cap B_r(y)|\ge3\}$, and for $r\ge2$ and $\pi\in\Pi$ let $\mathcal A_r(\pi)$ be the set of sites $A_r(c,\pi)$ such that $B_r(c)$ is $X$-paired with label $\pi$. By Lemma 8.3(2), $\mathcal A_r(\pi)\subseteq P$. \textbf{Lemma 8.6} (interior sites). \textit{Let $x=x_j\in P\cap B_K(v)$. Then $0\lambda_*r$, $|\Pi|(|\Pi|-1)\le|\Pi|^2$ and $(2h+1)^n\ge2^nh^n$, dividing by $k$ gives \[ \frac{(\delta-2)\,2^n}{|\Pi|}\,h^n\le2n\ell\Delta\,r\,a^{n-1}h^{n-1}+\frac{\Delta|\Pi|^2\,T\,a^n}{\lambda_*\,r}\,h^n. \tag{8.3} \] By the choice of $h$ the first term on the right is at most $\frac{(\delta-2)2^n}{2|\Pi|}h^n$, and by the choice of $r$ the second is strictly less than $\frac{(\delta-2)2^n}{2|\Pi|}h^n$, a contradiction.\qedsq \subsection*{8.6. Remarks and the case of $\mathbb Z^d$} \begin{enumerate} \item{} \textit{Use of $\delta\ge3$ and (X1)–(X4).} $\delta\ge3$ is used only in Lemma 8.8. The clause $A_\rho(c,\pi)\in S_\rho(c)$ of (X2) locates the apexes, the clause “every pair contains an apex” gives Lemma 8.3(1), and the injectivity is used at the two radii $\gamma h+1$ (Lemma 8.7) and $r$ (Lemma 8.10). (X3) and (X4) are used only in Lemma 8.5, at the single radius $r$. In particular (X4) needs to hold only for an unbounded set of radii, with $r$ replaced by the least such radius above the value in (8.1) and with $h$ and $K(\rho_0)$ then computed from this $r$ by (8.1), and data for one of the two orders of each pair of labels suffice. \item{} \textit{The degree condition.} It cannot be dropped from the argument in the form used here without a replacement, since it is the only source of the excess of $\mathcal W_r$ over $P$. On a periodic lattice, sites of degree $2$ occur in chains of bounded length $L_2$, and replacing each path site by the first site of degree at least $3$ that follows it on $P$ gives $|P\cap Q_H|\le(L_2+1)\Delta|(\mathcal W_r\setminus P)\cap Q_{H+\ell(L_2+1)}|$ in place of Lemma 8.8. The same proof then gives CL$_X$ without the hypothesis $\delta\ge3$, with a larger constant. \item{} \textit{Size of $K$.} $K$ is of order $2\ell\gamma^2h$ with $h$ and $r$ as in (8.1). The constants are not optimized. \end{enumerate} We now check that Theorem C contains the Corner Lemma CL($d$) of Section 2.7. Regard $\mathbb Z^d$ as a periodic lattice with $n=d$, $k=1$ and $x\sim y$ iff $\|x-y\|_1=1$. Then $d(x,y)=\|x-y\|_1$, $\Delta=\delta=2d$, $\ell=1$ and $\gamma=d$. Let $\mathcal U=\{\pm e_1,\dots,\pm e_d\}$, write $\pi\perp\eta$ if $\eta\notin\{\pm\pi\}$, and let \[ X^{\rm cor}_\rho(c)=\bigl\{\{c+\rho\pi,\ c+(\rho-1)\pi+\eta\}:\ \pi,\eta\in\mathcal U,\ \pi\perp\eta\bigr\}\qquad(\rho\ge2), \] the corner pairs of Definition 2.11. Then CL$_{X^{\rm cor}}$ is, word for word, the Corner Lemma CL($d$) of Section 2.7. \textbf{Lemma 8.14.} \textit{$X^{\rm cor}$ satisfies (X1)–(X4) with $\Pi=\mathcal U$, $A_\rho(c,\pi)=c+\rho\pi$, $\rho_X=2$, $p_0=2$, and the data $u=\pi_2-\pi_1$, $t_0=2$, $\lambda=1$ if $\pi_1\perp\pi_2$, and $u=-\pi_1$, $t_0=3$, $\lambda=1$ if $\pi_2=-\pi_1$. Moreover $N_\rho(c,\pi)=\{c+(\rho-1)\pi+\eta:\eta\perp\pi\}$.} \textit{Proof.} (X1) and (X2) are clear. For (X3): the vector $(\rho-1)\pi'+\eta'$ has two nonzero coordinates while $\rho\pi$ has one, so a corner pair containing $c+\rho\pi$ has apex $c+\rho\pi$ and partner $c+(\rho-1)\pi+\eta$ with $\eta\perp\pi$, at distance $\|\eta-\pi\|_1=2$. For (X4), write $N_j=A_j-\pi_j+\eta_j$ with $\eta_j\perp\pi_j$, and let $t_0\le t\le\rho$. \textit{Perpendicular poles:} $d(A_1,A_2)=2t\ge4$; $A_2-c_1=(\rho-t)\pi_1+t\pi_2$ has norm $\rho$; $N_2-c_1=(\rho-t)\pi_1+(t-1)\pi_2+\eta_2$ has norm $|\rho-t-1|+t-1\le\rho$ if $\eta_2=-\pi_1$, and $\rho$ otherwise; the other direction is symmetric. \textit{Antipodal poles} ($\pi_1=\sigma$, $u=-\sigma$): $d(A_1,A_2)=t\ge3$; $A_2-c_1=(\rho-t)\sigma$ and $N_2-c_1=(\rho-t+1)\sigma+\eta_2$ have norms $\rho-t$ and $\rho-t+2$, at most $\rho$; the other direction is symmetric.\qedsq For antipodal poles the data differ from a literal transcription of Lemma 4.5. The direction must be $u=-\pi_1$ (with $u=\pi_1$ the apex $A_2$ leaves $B_\rho(c_1)$), and $t_0=2$ fails because then $d(A_1,A_2)=2=p_0$: for $d=2$, $\rho=5$, $c_1=0$, $A_1=(5,0)$, $A_2=(3,0)$ and $c_2=(8,0)$ the two port pairs $\{(5,0),(4,1)\}$ and $\{(3,0),(4,1)\}$ share the partner $(4,1)$. The proof of Lemma 4.5 reaches $t=2$ by a finer argument that (X4) does not encode. \textbf{Corollary 8.15} (Theorem 4.8 with an explicit constant). \textit{Let $d\ge2$ and $\rho_0\ge2$. Then CL($d$) holds with} \[ r=\max\Bigl\{\rho_0,\Bigl\lfloor\frac{48d^4(4d+9)^d}{(d-1)2^d}\Bigr\rfloor+1\Bigr\},\quad h=\max\Bigl\{r,\Bigl\lceil\frac{8d^3(4d+9)^{d-1}r}{(d-1)2^d}\Bigr\rceil\Bigr\},\quad K=2d(dh+1)+r+1 . \] \textit{For $d=2$ and $\rho_0\le55\,489$ this gives $r=55\,489$, $h=15\,093\,008$ and $K=120\,799\,558$.} \textit{Proof.} Apply Theorem C with Lemma 8.14: $|\Pi|=2d$, $T=3$, $U=1$, $\lambda_*=1$, so $a=4d+9$, and (8.1) gives the stated values. For $d=2$: $48\cdot16\cdot17^2/4=55\,488$, $h=272r$ and $K=4(2h+1)+r+1$.\qedsq \section*{9. The honeycomb lattice} Recall that HEX has the sites $A(z)=(z,0)$ and $B(z)=(z,1)$, $z\in\mathbb Z^2$, and the edges $A(z)\sim B(w)$ for $z-w\in E:=\{0,e_1,e_2\}$. The root is $o=A(0)$, and we call $|v|=d(o,v)$ the \textit{level} of $v$. For a site $c$ let \[ \resizebox{473.2pt}{!}{$\displaystyle X^{\rm hex}_\rho(c)=\bigl\{\{C,N\}:\ C,N\in S_\rho(c),\ d(C,N)=2,\ C\text{ has exactly one neighbour }C'\text{ in }S_{\rho-1}(c),\text{ and }C'\text{ has a neighbour in }S_\rho(c)\text{ other than }C\bigr\}.$} \] A site $C$ with the property in this definition is called a \textit{corner} of $S_\rho(c)$. \textbf{Theorem B-HEX.} \textit{The honeycomb lattice satisfies (B1) with $M(\mathrm{HEX})=\{m\ge3\}$, (B2) with $M'(\mathrm{HEX})=\{n\ge1\}$, $n_0=1$, $t_0=3$, $c_2=1$, (B3) with $c_3=43$, $c_4=50$, $K_3=1$, and TP$_{X^{\rm hex}}$ with $\rho_{\rm TP}=200$ and $C_0=30$; and $X^{\rm hex}$ satisfies (X1)–(X4) with $\rho_X=2$ and $p_0=2$. Moreover, for $\rho\ge\max(200,m+30)$, every $c'\in B_1(c)$ and every escapable pair of $S_m(c')$, the pairs of distinct, non-adjacent, non-pocket sites of $S_\rho(c)$ that admit no design are exactly the twelve pairs of $X^{\rm hex}_\rho(c)$.} The twelve exceptional pairs sit at the six corners of the hexagonal ball: each corner forms an exceptional pair with each of its two neighbours along the boundary polygon of the sphere (Lemma 9.4). They admit no design for the reason found in [BBR, §2]: the corner has a unique inner neighbour, and every way into the ball from the corner comes too close to the partner (Proposition 9.9). Two layers do not suffice in (B2): $S_q\cup S_{q+1}$ is disconnected for every $q\ge1$ (see after Proposition 9.8). The honeycomb lattice is the main new case of Theorem 1.2. Theorem B-HEX, Theorem C, Theorem A and Proposition 6.6 give Conjecture 1 for site percolation on it (Section 12.11). The computer enters only through the three local lemmas of Section 9.5, at finitely many radii, and a transfer lemma extends them to all radii (Lemma 9.17; Section 13.2). \subsection*{9.1. Cube coordinates, distances and symmetries} For $x\in\mathbb Z^3$ write $s(x)=x_1+x_2+x_3$ and $\|x\|_1=|x_1|+|x_2|+|x_3|$, and let $f_1,f_2,f_3$ be the unit vectors of $\mathbb Z^3$. For a site $v=((z_1,z_2),i)$ put $q(v)=(-z_1,-z_2,z_1+z_2+i)$, let $i(v)=i$, and let $\sigma(v)=1-2i(v)\in\{\pm1\}$. \textbf{Lemma 9.1} (cube coordinates; distance formula). \textit{(a) $q$ is a bijection from $V$ onto $\{x\in\mathbb Z^3:s(x)\in\{0,1\}\}$ with $s(q(v))=i(v)$, and $q(\tau_yv)=q(v)+T(y)$ with $T(y)=(-y_1,-y_2,y_1+y_2)$. (b) The three neighbours of $v$ are the sites $v^{(j)}$ with $q(v^{(j)})=q(v)+\sigma(v)f_j$, $j=1,2,3$. (c) $d(v,v')=\|q(v)-q(v')\|_1$; that is, for $v=((a,b),i)$ and $v'=((a',b'),i')$,} \[ d(v,v')=|a-a'|+|b-b'|+|(a+b+i)-(a'+b'+i')|. \] \textit{(d) With $w=q(v)-q(c)$ and $\varepsilon=s(w)\in\{-1,0,1\}$ we have $d(c,v)=\max_j|2w_j-\varepsilon|$.} \textit{Proof.} (a) is immediate. (b) $q((w-\nu,1))-q((w,0))$ equals $f_1,f_2,f_3$ for $\nu=e_1,e_2,0$. (c) Images of adjacent sites differ by a unit vector, so $\|q(v)-q(v')\|_1\le d(v,v')$. Conversely, let $x=q(v)\ne y=q(v')$. If $i(v)=0$ and $y_j\le x_j$ for all $j$, then $s(x-y)=-s(y)\le0$ forces $x=y$; so some $y_j>x_j$, and the neighbour with image $x+f_j$ is closer to $y$. The case $i(v)=1$ is symmetric. Induction on $\|x-y\|_1$ gives (c). (d) $\|w\|_1=\max_{e\in\{\pm1\}^3}\langle e,w\rangle=\max(|\varepsilon|,\max_j|\varepsilon-2w_j|)$, and $|2w_j-\varepsilon|\ge|\varepsilon|$ by parity.\qedsq With $\lambda_1(x,y)=x$, $\lambda_2(x,y)=y$, $\lambda_3(x,y)=x+y$, put $h(z)=\max_l|\lambda_l(z)|$ and $g(z)=\max_l\max(\lambda_l(z),-\lambda_l(z)-1)$. By Lemma 9.1(d), applied with $c=o$ and $\varepsilon=0$, resp. $\varepsilon=1$, and the identity $|2t+1|=2\max(t,-t-1)+1$, \[ |A(z)|=2h(z),\qquad|B(z)|=2g(z)+1. \tag{9.1} \] Let $U_6=\{\pm e_1,\pm e_2,\pm(e_1-e_2)\}$. \textbf{Lemma 9.2} (automorphisms). \textit{The following maps are automorphisms of HEX: the translations $\tau_w$; $\iota(z,i)=(-z,1-i)$; $\sigma_0(z,i)=((z_2,z_1),i)$; and $\varrho(A(z))=A(Mz)$, $\varrho(B(z))=B(Mz-e_1)$, where $M(x,y)=(-x-y,x)$. The maps $\sigma_0$ and $\varrho$ fix $o$, $\varrho^3=\mathrm{id}$, and the translations together with $\iota$ act transitively on $V$.} \textit{Proof.} For $\iota$, $A(-w)\sim B(-z)$ iff $z-w\in E$. For $\sigma_0$, $E$ is invariant under exchanging coordinates. For $\varrho$, the affine map $t\mapsto Mt+e_1$ permutes $E$ ($0\mapsto e_1\mapsto e_2\mapsto0$), and $z-w\in E$ iff $Mz-(Mw-e_1)\in E$. Finally $M^3=I$ and $(M^2+M+I)e_1=0$.\qedsq \textbf{Corollary 9.3.} \textit{(a) HEX is bipartite with classes $\{A(z)\}$ and $\{B(z)\}$; along every edge the distance to a fixed site changes by exactly one; all sites of a sphere have the same type, and no two sites of a sphere are adjacent. (b) Two distinct sites have at most one common neighbour. (c) $d(A(z),A(z'))=d(B(z),B(z'))=2h(z-z')$; two sites of the same type at distance $2$ differ by a vector of $U_6$ and have exactly one common neighbour.} \textit{Proof.} (a) follows from Lemma 9.1 and $s(q(v))=i(v)$. (b) Common neighbours of $A(z)\ne A(z')$ are the $B(w)$ with $w\in(z-E)\cap(z'-E)$, and the six differences $e-e'$ ($e\ne e'\in E$) are distinct; similarly for B-sites. (c) follows from Lemma 9.1(c) and (b).\qedsq \subsection*{9.2. Spheres around the root} \textbf{Lemma 9.4} (shape of the spheres). \textit{Let $k\ge1$ and $u_0=(-1,1)$, $u_1=(-1,0)$, $u_2=(0,-1)$, $u_3=(1,-1)$, $u_4=(1,0)$, $u_5=(0,1)$ (indices mod $6$).} \textit{(a) $S_{2k}=\{A(z):z\in\partial H_k\}$, where $\partial H_k$ is the closed lattice polygon with vertices $a_0=(k,0)$, $a_1=(0,k)$, $a_2=(-k,k)$, $a_3=(-k,0)$, $a_4=(0,-k)$, $a_5=(k,-k)$ and sides $\{a_j+tu_j:0\le t\le k\}$. $|S_{2k}|=6k$.} \textit{(b) $S_{2k+1}=\{B(z):z\in\partial G_k\}$, where $\partial G_k$ has vertices $b_0=(k,0)$, $b_1=(0,k)$, $b_2=(-k-1,k)$, $b_3=(-k-1,0)$, $b_4=(0,-k-1)$, $b_5=(k,-k-1)$ and sides $\{b_j+tu_j:0\le t\le\ell_j\}$, with $\ell_j=k$ for even $j$ and $\ell_j=k+1$ for odd $j$. $|S_{2k+1}|=6k+3$.} \textit{(c) Vertices of the polygons have two outer neighbours and one inner neighbour. Interior points of a side $j$ of $\partial H_k$ have one outer and two inner neighbours if $j$ is even, and two outer and one inner if $j$ is odd; for $\partial G_k$ it is the other way round. In particular no sphere $S_n$, $n\ge1$, has a pocket.} \textit{(d) The corners of $S_{2k}$ are the six sites $A(a_j)$, and the corners of $S_{2k+1}$ are the six sites $B(b_j)$.} \textit{(e) For $n\ge4$ the set $X^{\rm hex}_n(o)$ consists of the twelve pairs formed by a corner and one of its two neighbours along the polygon: $\{A(a_j),A(a_j+u_j)\}$, $\{A(a_j),A(a_j-u_{j-1})\}$ for $n=2k$, and $\{B(b_j),B(b_j+u_j)\}$, $\{B(b_j),B(b_j-u_{j-1})\}$ for $n=2k+1$.} \textit{Proof.} For (a) and (b), by (9.1), $\{h\le k\}$ is the hexagon $|x|,|y|,|x+y|\le k$ and $\{g\le k\}$ is given by $x,y,x+y\le k$ and $x,y,x+y\ge-k-1$; their boundaries are the stated polygons. (c) For $A(z)\in S_{2k}$ let $J^\pm=\{l:\lambda_l(z)=\pm k\}$. Since $g(z-e)$ is the maximum of $\lambda_l(z)-\lambda_l(e)$ and $-\lambda_l(z)+\lambda_l(e)-1$, the neighbour $B(z-e)$ lies in $S_{2k+1}$ iff some $l\in J^+$ has $\lambda_l(e)=0$ or some $l\in J^-$ has $\lambda_l(e)=1$, and in $S_{2k-1}$ otherwise. Similarly, for $B(z)\in S_{2k+1}$ with $K^+=\{l:\lambda_l(z)=k\}$, $K^-=\{l:\lambda_l(z)=-k-1\}$, the neighbour $A(z+e)$ lies in $S_{2k+2}$ iff some $l\in K^+$ has $\lambda_l(e)=1$ or some $l\in K^-$ has $\lambda_l(e)=0$. We use $(\lambda_l(e))_l=(0,0,0),(1,0,1),(0,1,1)$ for $e=0,e_1,e_2$. At the vertices $a_0,\dots,a_5$ the pair $(J^+,J^-)$ is $(\{1,3\},\emptyset)$, $(\{2,3\},\emptyset)$, $(\{2\},\{1\})$, $(\emptyset,\{1,3\})$, $(\emptyset,\{2,3\})$, $(\{1\},\{2\})$, and in the interior of the sides $0,\dots,5$ it is $(\{3\},\emptyset)$, $(\{2\},\emptyset)$, $(\emptyset,\{1\})$, $(\emptyset,\{3\})$, $(\emptyset,\{2\})$, $(\{1\},\emptyset)$; the same list gives $(K^+,K^-)$ at the $b_j$ and on the sides of $\partial G_k$. Hence the outer neighbours $B(z-e)$ of $A(z)\in S_{2k}$ are given by $e\in\{0,e_2\},\{0,e_1\},\{0,e_1\},\{e_1,e_2\},\{e_1,e_2\},\{0,e_2\}$ at $a_0,\dots,a_5$ and by $e\in\{0\},\{0,e_1\},\{e_1\},\{e_1,e_2\},\{e_2\},\{0,e_2\}$ on the sides $0,\dots,5$; the outer neighbours $A(z+e)$ of $B(z)\in S_{2k+1}$ are given by $e\in\{e_1,e_2\},\{e_1,e_2\},\{0,e_2\},\{0,e_2\},\{0,e_1\},\{0,e_1\}$ at $b_0,\dots,b_5$ and by $e\in\{e_1,e_2\},\{e_2\},\{0,e_2\},\{0\},\{0,e_1\},\{e_1\}$ on the sides $0,\dots,5$. Every site has three neighbours, in the two adjacent spheres, which gives the inner counts. For $S_1=\{B(0),B(-e_1),B(-e_2)\}$ each site has the inner neighbour $o$ and two outer neighbours. (d) By (c), the sites with exactly one inner neighbour are the vertices and the interior points of the odd sides of $\partial H_k$, resp. of the even sides of $\partial G_k$. At $A(a_0)$ the inner neighbour $B(k-1,0)$ is adjacent to $A(k-1,1)\in S_{2k}$, while at an interior point $A(k,y)$ of side $5$ ($-k2$ and that six explicit vectors lie in $\mathfrak B$, i.e. that each coordinate lies in $[L,H]$. The indices of $\pi_2$ and the vectors $U$, $R_{\pi_2}$ are read off from the table of labels. \textit{Case A: $(\pi,\pi+1)$, $\pi$ even; $u=\delta_{\pi+2}$, $t_0=2$, $\lambda=\frac12$.} Here $(h(\pi_2),l(\pi_2),m(\pi_2))=(h,m,l)$, $U=f_l-f_m$, $R_{\pi_2}=\{f_m-f_l,f_l-f_h\}$, and the six vectors are $(H,L+t,-t)$, $(H,L+t-1,1-t)$, $(H-1,L+t+1,-t)$, $(H,-t,L+t)$, $(H,1-t,L+t-1)$, $(H-1,-t,L+t+1)$, whose coordinates lie in $[L,H]$ by (E1), (E2), (E4) for $2\le t\le\rho/2$; and $d(A_1,A_2)=2t\ge4$. \textit{Case B: $(\pi,\pi+1)$, $\pi$ odd; same data.} Here $(h(\pi_2),l(\pi_2),m(\pi_2))=(m,l,h)$, $U=f_m-f_h$, $R_{\pi_2}=\{f_l-f_h,f_h-f_m\}$, and the vectors are $(H-t,L,t)$, $(H-t-1,L+1,t)$, $(H-t+1,L,t-1)$, $(t,L,H-t)$, $(t,L+1,H-t-1)$, $(t-1,L,H-t+1)$; $d(A_1,A_2)=2t\ge4$. \textit{Case C: $(\pi,\pi+2)$, $\pi$ even; $u=\delta_{\pi+2}-\delta_\pi$, $t_0=1$, $\lambda=\frac12$.} Here $(h(\pi_2),l(\pi_2),m(\pi_2))=(l,m,h)$, $U=-f_h+2f_l-f_m$, $R_{\pi_2}=\{f_m-f_h,f_h-f_l\}$, and the vectors are $(H-t,L+2t,-t)$, $(H-t-1,L+2t,1-t)$, $(H-t+1,L+2t-1,-t)$, $(t,H-2t,L+t)$, $(t,H-2t+1,L+t-1)$, $(t-1,H-2t,L+t+1)$, in $\mathfrak B$ by (E1)–(E3) for $1\le t\le\rho/2$; $d(A_1,A_2)=4t\ge4$. \textit{Case D: $(\pi,\pi+2)$, $\pi$ odd; same data.} Here $(h(\pi_2),l(\pi_2),m(\pi_2))=(m,h,l)$, $U=-2f_h+f_l+f_m$, $R_{\pi_2}=\{f_h-f_l,f_l-f_m\}$, and the vectors are $(H-2t,L+t,t)$, $(H-2t+1,L+t-1,t)$, $(H-2t,L+t+1,t-1)$, $(L+2t,-t,H-t)$, $(L+2t,1-t,H-t-1)$, $(L+2t-1,-t,H-t+1)$; $d(A_1,A_2)=4t\ge4$. \textit{Case E: $(\pi,\pi+3)$; $u=\delta_{\pi+3}$, $t_0=2$, $\lambda=1$.} Here $(h(\pi_2),l(\pi_2),m(\pi_2))=(l,h,m)$, $U=f_l-f_h$, $R_{\pi_2}=\{f_h-f_m,f_m-f_l\}$, and the vectors are $(H-t,L+t,0)$, $(H-t+1,L+t,-1)$, $(H-t,L+t-1,1)$, $(L+t,H-t,0)$, $(L+t,H-t+1,-1)$, $(L+t-1,H-t,1)$, in $\mathfrak B$ by (E4), (E5) for $2\le t\le\rho$; $d(A_1,A_2)=2t\ge4$. In all cases $d(A_1,A_2)>2=p_0$ and the containments hold, for every $\rho\ge2$.\qedsq \textbf{Remark 9.24.} The tables give the exact sets of $t\ge1$ for which the conclusions of (X4) hold with the chosen $u$: $[2,-L]$ in Case A, $[2,H]$ in Case B, $[1,\lfloor\rho/2\rfloor]$ in Cases C and D, and $[2,\rho]$ in Case E. So $\lambda=\frac12$ cannot be enlarged in Cases A and B, and $t_0=2$ cannot be lowered in Cases A, B and E (for $t=1$, $d(A_1,A_2)=2=p_0$). In Cases A and B the apex $A_2$ stays on the sphere $S_\rho(c_1)$, so the containment of the partners is a genuinely discrete fact. \subsection*{9.10. Proof of Theorem B-HEX, and the corner constant} \textit{Proof of Theorem B-HEX.} (B1) and (B2) follow from Propositions 9.7 and 9.8, and (B3) is proved in Section 9.7. TP$_{X^{\rm hex}}$ follows from Theorem 9.18, and (X1)–(X4) follow from Theorem 9.20; the statement on the exact set of failing pairs follows from Corollary 9.19.\qedsq \textit{The corner constant.} The constant of Theorem C for the honeycomb lattice is computed from $n=2$, $\Delta=\delta=3$, $|\Pi|=6$, $T=2$, $U=2$ and $\lambda_*=\frac12$ (Theorem 9.20(4)), and $\ell=1$, $\gamma=5$: by Lemma 9.1(c), sites in cells $z,z'$ with $\|z-z'\|_\infty\le1$ are at distance at most $1+1+3=5$, with equality for $A(0)$ and $B(1,1)$, and adjacent sites lie in cells at sup-distance at most $1$. Then (8.1) gives $a=31$, $r=1\,245\,457$, $h=1116\,r=1\,389\,930\,012$ and $K_{\rm CL}(\rho_0)=50h+r+11=69\,497\,746\,068$ for every $\rho_0\le1\,245\,457$. \section*{10. The Kagomé lattice and bond percolation on the honeycomb lattice} \textbf{Theorem B-KAG.} \textit{The Kagomé lattice satisfies (B1) with $M(\mathrm{KAG})=\{m\ge2:m\text{ even}\}$, (B2) with $M'(\mathrm{KAG})=\{n\ge2:n\text{ even}\}$, $n_0=2$, $t_0=3$, $c_2=2$, (B3) with $c_3=22$, $c_4=30$, $K_3=1$, and TP$_\emptyset$ with $\rho_{\rm TP}=8$ and $C_0=7$. Moreover, for every site $x$ and every odd $m=2k+1\ge3$ the sphere $S_m(x)$ has exactly two pockets, while $S_1(x)$ and the spheres of even radius have none.} The Kagomé lattice lies in the regime that [BBR] call much easier. Site percolation on KAG is bond percolation on the honeycomb lattice HEX: the bijection $\varepsilon$ of Lemma 10.1 identifies a set of open sites of KAG with a set of open edges of HEX. The combinatorial core of our proof of TP$_\emptyset$ (Theorem 10.10) is the argument outlined in [BBR, §6] for bond percolation on $\mathbb Z^d$, transported to KAG through the identification KAG $=L(\mathrm{HEX})$. Sets of sites of KAG that do not touch are edge sets of HEX with disjoint vertex sets (Lemma 10.2(a)), and the two required connections are obtained as vertex-disjoint paths from Menger’s theorem with a $2$-connected core. What is specific to KAG is the verification that this argument applies under the hypotheses of TP$_\emptyset$. The ports are arbitrary distinct, non-adjacent, non-pocket sites of $S_\rho(c)$, which are radial edges (Lemma 10.8). The gadget sphere is centred at an arbitrary site $c'$ with $d(c,c')\le1$, so that two level functions have to be controlled at the same time. The terminals form an arbitrary escapable pair. The attachments of the ports and of the terminals to the core rest on an explicit level function (Lemmas 10.3 and 10.4) and on a lemma that excludes funnels (Lemma 10.7). Finally, the boundary properties (B2) and (B3) are proved in Proposition 10.11(b) and Section 10.4. Section 10.5 deduces Corollary 1.3. \textit{Notation.} For $t=(x,y)\in\mathbb Z^2$ let $N(t)=\max(|x|,|y|,|x+y|)$, $s(t)=x+y$ and $\bar t=(y,x)$, and let $U_6=\{\pm e_1,\pm e_2,\pm(e_1-e_2)\}$; points of $\mathbb Z^2$ differing by an element of $U_6$ are \textit{T-adjacent}. Put $\Delta(w)=\{w,w+e_1,w+e_2\}$. In the honeycomb lattice HEX, $A(z)\sim B(w)$ iff $z\in\Delta(w)$; we write $d_H$ for its distance. $[\Phi]$ is the indicator of a statement $\Phi$. \subsection*{10.1. The line graph} \textbf{Lemma 10.1.} \textit{The map $\varepsilon(a(z))=\{A(z),B(z)\}$, $\varepsilon(b(z))=\{A(z+e_1),B(z)\}$, $\varepsilon(c(z))=\{A(z+e_2),B(z)\}$ is a bijection from the sites of KAG onto the edges of HEX that commutes with translations, and two distinct sites of KAG are adjacent if and only if their images share an endpoint. Thus KAG is the line graph of HEX.} \textit{Proof.} Every edge of HEX is $\{A(w+\eta),B(w)\}$ for a unique $w$ and $\eta\in\{0,e_1,e_2\}$. The three edges at $B(w)$ are the images of $a(w),b(w),c(w)$, and the three edges at $A(z)$ are the images of $a(z),b(z-e_1),c(z-e_2)$. So the pairs of sites whose images share an endpoint are the pairs inside up triangles and down triangles, which are the edges of KAG.\qedsq We identify a site $f$ with the edge $\varepsilon(f)$, and write $V(f)$ for its two endpoints and $V(F)=\bigcup_{f\in F}V(f)$. \textbf{Lemma 10.2.} \textit{(a) Two sets $F,F'$ of sites touch if and only if $V(F)\cap V(F')\ne\emptyset$. (b) A nonempty set $F$ of sites is connected if and only if the subgraph $(V(F),F)$ of HEX is connected. (c) For distinct sites, $d(f,f')=1+\min\{d_H(p,p'):p\in V(f),p'\in V(f')\}$. (d) A sequence $(f_0,\dots,f_q)$, $q\ge1$, is an induced path of KAG if and only if there are pairwise distinct vertices $p_0,\dots,p_{q+1}$ of HEX with $f_j=\{p_j,p_{j+1}\}$.} \textit{Proof.} (a) and (b) follow from Lemma 10.1. (c) Common endpoints of consecutive sites of a path $f=f_0,\dots,f_k=f'$ form a walk in HEX of length at most $k-1$; conversely a geodesic $r_0,\dots,r_\ell$ of HEX from $V(f)$ to $V(f')$ gives the walk $f,\{r_0,r_1\},\dots,\{r_{\ell-1},r_\ell\},f'$. (d) If the $p_j$ are distinct, sites at index distance at least $2$ have disjoint vertex sets. Conversely, in an induced path let $p_j$ be the common endpoint of $f_{j-1}$ and $f_j$; $p_j=p_{j+1}$ would make $f_{j-1}$ and $f_{j+1}$ adjacent, and $p_i=p_j$ with $i0]$ and $D(B(z))=2N(z)-[s(z)<0]$.} \textit{Proof.} Each element of $U_6$ changes $x$, $y$ and $x+y$ by at most one, and $N(t)$ steps suffice (for $x,y\ge0$ use $e_1,e_2$; for $x\le0\le y$ use $(-1,1)$ together with $e_2$ or $-e_1$; the rest by symmetry). Two A-vertices at distance two along a path, $A(t)\sim B(w)\sim A(t')$, satisfy $t'-t\in U_6\cup\{0\}$, and every T-adjacent pair lies in a common triangle $\Delta(w)$; this gives the A–A formula, and $\iota$ gives the B–B formula. The neighbours of $B(w)$ are the $A(t)$, $t\in\Delta(w)$. For the levels, $D(A(z))=\min\bigl(2N(z),1+2N(z-e_1),1+2N(z-e_2)\bigr)$, so it suffices to show that $\min(N(z-e_1),N(z-e_2))\le N(z)-1$ if and only if $s(z)>0$. Let $n=N(z)$, $z=(x,y)$, $s=x+y$. If $s\ge1$ and $x\ge1$, then $|y|\le n-1$, $0\le x-1\le n-1$ and $0\le s-1\le n-1$, so $N(z-e_1)\le n-1$; if $s\ge1$ and $x\le0$, then $y\ge1$ and symmetrically $N(z-e_2)\le n-1$. If $s\le0$, we show $N(z-e_1),N(z-e_2)\ge n$, using that one of $|x|,|y|,|s|$ equals $n$. If $|s|=n$, then $|s-1|=n+1$. If $x=-n$, then $|x-1|=n+1$ and $N(z-e_2)\ge|x|=n$. If $x=n>0$, then $y\le-x$ forces $y=-n$, so $N(z-e_1)\ge|y|=n$ and $|y-1|=n+1$. The cases $y=\pm n$ are symmetric. The formula for $B(z)$ follows from $D\circ\iota=D$ and $\iota(A(-z))=B(z)$.\qedsq \textbf{Lemma 10.4} (sign rule). \textit{With respect to $o$, and for $z=(x,y)$: $B(z-e_1)$ is an inner neighbour of $A(z)$ if $x>0$ and an outer one if $x\le0$; $B(z-e_2)$ is inner if $y>0$ and outer if $y\le0$; $B(z)$ is inner, tangential or outer according as $s(z)<0$, $=0$, $>0$. Equivalently, for $w=(x,y)$: $A(w+e_1)$ is inner for $B(w)$ iff $x<0$; $A(w+e_2)$ is inner iff $y<0$; $A(w)$ is inner, tangential or outer according as $s(w)>0$, $=0$, $<0$.} \textit{Proof.} The statement about $B(z)$ is Lemma 10.3. For $B(z-e_1)$, $\Delta:=D(B(z-e_1))-D(A(z))=2(N(z-e_1)-N(z))+1-2[s\le0]$ is odd, so $\Delta=\pm1$. Let $n=N(z)$. If $s\ge1$: for $x\ge1$, $N(z-e_1)=n-1$ by the proof of Lemma 10.3, so $\Delta=-1$; for $x\le0$, $n=\max(|x|,y)\le N(z-e_1)$, so $\Delta=1$. If $s\le0$: $N(z-e_1)-n\in\{0,1\}$, so $\Delta=-1$ iff $N(z-e_1)\ne n+1$, and since the term $|y|$ does not change, $N(z-e_1)=n+1$ iff $x=-n$ or $s=-n$. For $x\ge1$ neither holds ($s=-n$ would give $y<-n$), so $\Delta=-1$. For $x\le0$ one has $n=-x$ if $y\ge0$ (as $|y|,|s|\le-x$) and $n=-s$ if $y<0$ (as $|x|,|y|\le|s|$), so $\Delta=1$. The statement for $B(z-e_2)$ follows by $\sigma$, and the second form is the first applied to the three edges at $B(w)$.\qedsq \textbf{Corollary 10.5} (distance formula). \textit{For $z=(x,y)$: $|a(z)|=2N(z)+[x+y=0\ne x]$, $|b(z)|=\max(|2x+1|,2|y|,|2x+2y+1|)$ and $|c(z)|=\max(2|x|,|2y+1|,|2x+2y+1|)$.} \textit{Proof.} By (10.1), $|f|=1+\min_{p\in V(f)}D(p)$ for $f\ne o$. For $a(z)$ use Lemma 10.3. For $b(z)=\{A(z+e_1),B(z)\}$, the inner endpoint is $A(z+e_1)$ iff $x<0$ (Lemma 10.4). With $Q=\max(|2x+1|,2|y|,|2s+1|)$: if $x\ge0$, $s\ge0$, then $|b(z)|=1+2N(z)$, which equals $Q$ because $|y|\le\max(x,s)$; if $x\ge0>s$, then $N(z)=|y|\ge\max(x+1,|s|)$ and $|b(z)|=2|y|=Q$; if $x<0\le s$, then $|b(z)|=2N(z+e_1)=2y=Q$; if $x<0$, $s<0$, then $|b(z)|=1+2N(z+e_1)=\max(-2x-1,2|y|+1,-2s-1)=Q$, because $|y|\ge-x$ and $|y|\ge-s$ cannot both hold. Finally $\sigma$ maps $b(z)$ to $c(\bar z)$ and fixes $o$.\qedsq \textbf{Corollary 10.6.} \textit{(a) Every vertex $p$ has an outer neighbour, and if $D_c(p)\ge1$ it has one or two inner neighbours. (b) For $c=o$ the tangential sites are exactly the sites $a(k,-k)$, $k\in\mathbb Z$, both of whose endpoints have level $2|k|$.} \textit{Proof.} Take $c=o$. By Lemma 10.4, $A(z)$ has no outer neighbour iff $x>0$, $y>0$ and $s(z)\le0$, which is impossible, and similarly for $B(w)$. A vertex at level at least $1$ has an inner neighbour on a geodesic, and at most two since it has an outer one. (b) is the case $s(z)=0$ of Lemma 10.4.\qedsq \textbf{Lemma 10.7} (no funnels). \textit{(a) No vertex $p$ with $D_c(p)\ge2$ has two distinct outer neighbours $r,r'$ with $\mathrm{IN}_c(r)=\mathrm{IN}_c(r')=\{p\}$. (b) No vertex $p$ has two distinct inner neighbours $r,r'$ with $\mathrm{OUT}_c(r)=\mathrm{OUT}_c(r')=\{p\}$.} \textit{Proof.} Take $c=o$ and, applying $\iota$, $p=A(z)$, $z=(x,y)$, $s=x+y$. (a) By Lemma 10.4: $B(z-e_1)$ is outer iff $x\le0$, and then $\mathrm{IN}(B(z-e_1))=\{A(z)\}$ iff $y\ge0$ and $s\le1$; $B(z-e_2)$ is outer iff $y\le0$, and then $\mathrm{IN}=\{A(z)\}$ iff $x\ge0$ and $s\le1$; $B(z)$ is outer iff $s>0$, and then $\mathrm{IN}=\{A(z)\}$ iff $x,y\ge0$. Two of these three conditions hold together only for $z=0$, $(0,1)$ or $(1,0)$, at levels $0$ and $1$. (b) $B(z-e_1)$ is inner iff $x>0$, and then $\mathrm{OUT}=\{A(z)\}$ iff $y<0$ and $s\ge1$; $B(z-e_2)$ is inner iff $y>0$, and then $\mathrm{OUT}=\{A(z)\}$ iff $x<0$ and $s\ge1$; $B(z)$ is inner iff $s<0$, and then $\mathrm{OUT}=\{A(z)\}$ iff $x,y<0$. No two of these hold together.\qedsq \textbf{Lemma 10.8} (pockets). \textit{Let $\rho\ge1$ and $f\in S_\rho(c)$. Then $\min_{p\in V(f)}D_c(p)=\rho-1$, and $f$ is a pocket of $S_\rho(c)$ if and only if it is tangential. So every non-pocket $f\in S_\rho(c)$ is radial with $D_c(f^{\rm in})=\rho-1$ and $D_c(f^{\rm out})=\rho$. The sphere $S_\rho(c)$ has no pocket if $\rho$ is even or $\rho=1$, and exactly two if $\rho=2k+1\ge3$: $a(z\pm(k,-k))$ for $c=a(z)$, $b(z\pm(0,k))$ for $c=b(z)$, and $c(z\pm(k,0))$ for $c=c(z)$.} \textit{Proof.} The first claim is (10.1). If $f$ is tangential, every site sharing an endpoint with $f$ has an endpoint at level $\le\rho-1$, hence lies in $B_\rho(c)$. If $f$ is radial, $f^{\rm out}$ has an outer neighbour $r$ (Corollary 10.6), and $\{f^{\rm out},r\}\in S_{\rho+1}(c)$ is adjacent to $f$. For $c=o$, Corollary 10.6(b) gives the pockets; for $c=(z,i)$ apply $\tau_z\varphi^i$, using $\varphi(a(t))=b(Rt)$, $\varphi(b(t))=c(Rt)$, $R(\pm(k,-k))=\pm(0,k)$ and $R(\pm(0,k))=\pm(-k,0)$.\qedsq \subsection*{10.3. Rings and the two-port property} For $n\ge1$ let $H_n=\{t:N(t)=n\}$, listed cyclically as $t_0=(n,0)$, $t_{j+1}=t_j+d_j$ (indices mod $6n$), where $d_j$ equals $(-1,1)$, $(-1,0)$, $(0,-1)$, $(1,-1)$, $(1,0)$, $(0,1)$ for $j$ in the six consecutive blocks of length $n$. \textbf{Lemma 10.9} (rings). \textit{Let $n\ge1$. (a) Two points of $H_n$ are T-adjacent iff they are cyclically consecutive. (b) For each $j$ there is a unique $w_j$ with $\{t_j,t_{j+1}\}\subseteq\Delta(w_j)$, and $\Gamma_n=(A(t_0),B(w_0),A(t_1),\dots,A(t_{6n-1}),B(w_{6n-1}))$ is a cycle of HEX of length $12n$. (c) The vertex sets of the $\Gamma_n$ are pairwise disjoint and cover all vertices except $A(0)$. (d) With $\mathrm{ind}(A(0))=0$ and $\mathrm{ind}(p)=n$ for $p\in\Gamma_n$, adjacent vertices have indices differing by at most one. (e) Every $p\in\Gamma_n$ satisfies $2n-2\le D(p)\le2n+1$.} \textit{Proof.} (a) The linear maps $t\mapsto-t$, $t\mapsto\bar t$ and $(x,y)\mapsto(-y,x+y)$ preserve $N$ and $U_6$ and permute the corners and sides of the hexagon, so it suffices to check the corner $(0,n)$ and a point $(x,n)$ with $-n0$; $B(z-e_2)$ is inner iff $Y>0$; $B(z)$ is inner, tangential, outer according as $X+Y<0$, $=0$, $>0$; and with $(X',Y')=w-q$, $A(w+e_1)$ is inner for $B(w)$ iff $X'<0$, $A(w+e_2)$ iff $Y'<0$, and $A(w)$ is inner, tangential, outer according as $X'+Y'>0$, $=0$, $<0$. (b) With respect to $b(q)$ and $(X,Y)=z-q$: $B(z-e_2)$ is inner for $A(z)$ iff $Y>0$; $B(z)$ is inner iff $X+Y\le0$ (outer otherwise); $B(z-e_1)$ is inner iff $X\ge2$, tangential iff $X=1$, outer iff $X\le0$; and with $(X',Y')=w-q$, $A(w+e_1)$ is inner for $B(w)$ iff $X'\le-1$, tangential iff $X'=0$, outer iff $X'\ge1$; $A(w+e_2)$ is inner iff $Y'\le-1$; $A(w)$ is inner iff $X'+Y'\ge1$ (outer otherwise).} \textit{Proof.} (a) follows from $D_{a(q)}=D\circ\tau_{-q}$ and Lemma 10.4. (b) $\tau_q\varphi$ maps $o$ to $b(q)$, and $(\tau_q\varphi)^{-1}(A(z))=A(\tilde z)$ with $\tilde z=(Y,1-X-Y)$, while $B(z)$, $B(z-e_1)$, $B(z-e_2)$ go to $B(\tilde z-e_2)$, $B(\tilde z)$, $B(\tilde z-e_1)$. Apply Lemma 10.4 at $A(\tilde z)$; the second part is the first applied to the three edges at $B(w)$.\qedsq For $q\in\mathbb Z^2$ define the site $w(q)$ by $w(q)=b(q)$ if $q_x,q_y\ge0$; $w(q)=a(q)$ if $q_x<0\le q_y$; and $w(q)=\iota(w(-q))$ if $q_y<0$ (so that $w(q)=b(q-e_1)$ if $-q_x,-q_y\ge0$ and $w(q)=a(q)$ otherwise). \textbf{Lemma 10.14.} \textit{Let $w=w(q)$. (a) $d(a(q),w)\le1$, both endpoints $r$ of $w$ satisfy $N(\pi(r))\ge N(q)-1$, and $N(q)\le(|w|+2)/2$. (b) For each endpoint $r$ of $w$ there is an infinite path $r=\zeta_0,\zeta_1,\dots$ of HEX avoiding the other endpoint, with $D(\zeta_k)=D(r)+k$ and $D_w(\zeta_k)=k$. (c) Every vertex $p$ with $N(\pi(p))\ge N(q)+2$ has a neighbour $p'$ with $D(p')=D(p)-1$ and $D_w(p')=D_w(p)-1$.} \textit{Proof.} (a) $a(q)$ and $b(q)$ share $B(q)$, and $a(q)$, $b(q-e_1)$ share $A(q)$; the endpoints of $w$ have coordinates within $N$-distance $1$ of $q$, so $D(r)\ge2N(q)-3$ and $|w|=1+\min_rD(r)$ gives the bound. For (b) and (c) it suffices, by $\iota$ (which fixes $o$), to treat $q_y\ge0$. (b) If $q_x,q_y\ge0$ and $w=b(q)$: from $B(q)$ take $\zeta_{2k}=B(q+ke_2)$, $\zeta_{2k+1}=A(q+(k+1)e_2)$, and from $A(q+e_1)$ take $\zeta_{2k}=A(q+(k+1)e_1)$, $\zeta_{2k+1}=B(q+(k+1)e_1)$; by Lemmas 10.4 and 10.13(b) each step is outer for $D$ and for $D_w$. If $q_x<0\le q_y$ and $w=a(q)$: from $A(q)$ take $\zeta_{2k}=A(q+(-k,k))$, $\zeta_{2k+1}=B(q+(-k-1,k))$, and from $B(q)$ take $\zeta_{2k}=B(q+(-k,k))$, $\zeta_{2k+1}=A(q+(-k,k+1))$; the steps $A(p)\to B(p-e_1)$ with $p_x\le q_x\le0$ and $B(r)\to A(r+e_2)$ with $r_y\ge q_y\ge0$ are outer for both by Lemmas 10.4 and 10.13(a). (c) If $w=a(q)$ with $q_x<0\le q_y$ and $p=A(t)$, a common inner neighbour exists if $t_x>0$, $t_y>q_y$ or $s(t)<\min(0,s(q))$; if none holds, then $N(t)\le N(q)$. For $p=B(t)$ a common inner neighbour exists if $t_x\max(0,s(q))$, and otherwise $N(t)\le N(q)$. If $w=b(q)$ with $q_x,q_y\ge0$: for $p=A(t)$ a common inner neighbour exists if $t_y>q_y$, $s(t)<0$ or $t_x\ge q_x+2$, and otherwise $N(t)\le N(q)+1$; for $p=B(t)$ if $t_x<0$, $t_y<0$ or $s(t)>s(q)$, and otherwise $N(t)\le N(q)$.\qedsq \textbf{Proposition 10.15.} \textit{(B3) holds for KAG with $c_3=22$, $c_4=30$ and $K_3=1$.} \textit{Proof.} Let $K\ge1$, $L\ge4K+120$ and $L-K-10\le|v|\le L$. Put $h=L-K-20$. A geodesic from $o$ to $v$ contains $w_0=(q,i)$ with $|w_0|=h$ and $d(w_0,v)\le K+20$. Let $w=w(q)$; as $a(q),b(q),c(q)$ are pairwise adjacent, Lemma 10.14(a) gives $d(w_0,w)\le2$, so $d(v,w)\le K+22$, $L-K-22\le|w|\le L-K-18$ (in particular $K+30<|w|0$ for $p\in(\pi(s),p_c(\mathrm{KAG}))$, and $\theta^{\rm bond}_F\ge\theta_E$ by Proposition 10.17(5).\qedsq \textit{Vertex activation.} Let $F^{\rm A},F^{\rm B}$ be rules attached to $A(0)$ and $B(0)$, of range $r$ in the sense that $F^{\rm X}(\eta)$ depends only on, and lies in, the set of edges with an endpoint within distance $r$ of the vertex. Such edges lie in $\mathfrak B_{r+1}(\varepsilon_0)$, resp. $\mathfrak B_{r+1}(\varepsilon_1)$. So $E_a=\Phi^{-1}F^{\rm A}\Phi$, $E_b=\Phi^{-1}F^{\rm B}\Phi$, $E_c\equiv\emptyset$ is an enhancement of site percolation on KAG with range $r+1$ (Definition 6.2 allows different rules on different orbits, including the empty rule). Assigning $A(z)$ to $a(z)$ and $B(z)$ to $b(z)$, a Bernoulli($s$) set of sites of KAG induces a Bernoulli($s$) set of vertices of HEX, independent of $\Phi\omega$, and the enhanced configurations correspond. By Lemma 10.16 essentialness corresponds, and Proposition 10.17(5),(6) apply unchanged. This proves Corollary 1.3 for vertex activation.\qedsq The crossing events of the two models do not correspond exactly, but they are not needed: Claim 5 and Theorem 6.5 are applied on KAG, and only $\theta$ and $p_c$ are transferred. The bond case outlined in [BBR, §6] uses the $2$-connectedness of the boundaries of $\ell^\infty$ boxes of $\mathbb Z^d$; the spheres of HEX are independent sets, so that outline does not apply to HEX as it stands. \section*{11. The triangular lattice} \textbf{Theorem B-TRI.} \textit{The triangular lattice satisfies (B1) with $M(\mathrm{TRI})=\{m\ge3\}$, (B2) with $M'(\mathrm{TRI})=\{n\ge1\}$, $n_0=1$, $t_0=1$, $c_2=1$, (B3) with $c_3=20$, $c_4=30$, $K_3=1$, and TP$_\emptyset$ with $\rho_{\rm TP}=16$ and $C_0=15$.} Theorem 1.2 for the triangular lattice is stated in [BBR, Theorem 3], and its proof is outlined in [BBR, §6]. We claim nothing new here. We include a complete proof because it is the simplest instance of the method: there are no exceptional pairs, and TP$_\emptyset$ follows the pattern of Section 3 (a local escape at the ports, the first step of [BBR, §6]; a step that moves arbitrary terminals outward and separates them; and a routing between two rings). Together with Theorem A and Proposition 6.6, Theorem B-TRI gives Theorem 1.2 for the triangular lattice (Section 12.11). Since $\kappa(\mathrm{TRI})=0$, the gadget centre is $c'=c$. We identify $V$ with $\mathbb Z^2$, so that $z\sim w$ iff $z-w\in U=\{\pm(1,0),\pm(0,1),\pm(1,-1)\}$. Two sites whose norms differ by at least $2$ are neither equal nor adjacent; we call this the \textit{layer rule}. For $x\in S_k$ let $\mathrm{in}(x)$ and $\mathrm{out}(x)$ be its sets of neighbours in $S_{k-1}$ and $S_{k+1}$. \subsection*{11.1. Cones, distances and spheres} \textit{Generators and cones.} Let $\xi_0=(1,0)$, $\xi_1=(0,1)$, $\xi_2=(-1,1)$, $\xi_3=(-1,0)$, $\xi_4=(0,-1)$, $\xi_5=(1,-1)$, indices modulo $6$. Then $U=\{\xi_0,\dots,\xi_5\}$, \[ \xi_{i+3}=-\xi_i,\qquad\xi_{i+1}=\xi_i+\xi_{i+2}, \tag{11.1} \] and each pair $(\xi_i,\xi_{i+1})$ is a basis of $\mathbb Z^2$. The linear map $\phi(z_1,z_2)=(-z_2,z_1+z_2)$ satisfies $\phi(\xi_i)=\xi_{i+1}$, so it is an automorphism fixing $o$; so are the reflection $(z_1,z_2)\mapsto(z_2,z_1)$ and the translations. Let $f_0(z)=z_1+z_2$, $f_1(z)=z_2$, $f_2(z)=-z_1$ and $f_{i+3}=-f_i$. Then $f_i=f_0\circ\phi^{-i}$ and \[ f_i(\xi_i)=f_i(\xi_{i+1})=1,\qquad f_i(\xi_{i+2})=f_i(\xi_{i+5})=0,\qquad f_i(\xi_{i+3})=f_i(\xi_{i+4})=-1. \tag{11.2} \] The \textit{cone} $\Sigma_i$ is the set of sites $\alpha\xi_i+\beta\xi_{i+1}$ with integers $\alpha,\beta\ge0$. \textbf{Lemma 11.1} (distance formula). \textit{For all $z,w$ and $i$: (1) $|f_i(z)-f_i(w)|\le d(z,w)$; (2) every site lies in some cone, and if $z=\alpha\xi_i+\beta\xi_{i+1}$ with $\alpha,\beta\ge0$ then $|z|=\alpha+\beta=f_i(z)$; (3) $|z|=\max_jf_j(z)=\max(|z_1|,|z_2|,|z_1+z_2|)$.} \textit{Proof.} (1) By (11.2), $f_i$ changes by at most one along an edge. (2) If $z_1,z_2\ge0$ then $z=z_1\xi_0+z_2\xi_1$; if $z_1\le0\le z_1+z_2$ then $z=(z_1+z_2)\xi_1+(-z_1)\xi_2$; if $z_1+z_2\le0\le z_2$ then $z=z_2\xi_2+(-z_1-z_2)\xi_3$; the other three sign patterns follow by $z\mapsto-z$. A walk of $\alpha$ steps $\xi_i$ and $\beta$ steps $\xi_{i+1}$ gives $|z|\le\alpha+\beta$, and (1) gives $|z|\ge f_i(z)=\alpha+\beta$. (3) follows from (1) and (2).\qedsq Consequently, for fixed $z,h$ the function $t\mapsto|z+th|=\max_j(f_j(z)+tf_j(h))$ is convex, and on an integer interval it attains its maximum at an end. \textbf{Lemma 11.2} (spheres are hexagonal cycles). \textit{Let $k\ge1$. For $s\in\mathbb Z$ write $s=ik+j$ with $0\le jk$ then $D\ge d-d/k-1\ge(k+1)(k-1)/k-1>1$. Hence $a_1$ and $g_1$ are distinct and non-adjacent. By Lemma 11.3, $\mathrm{in}(g)\subseteq\{P_{k-1}(q+D-1),P_{k-1}(q+D)\}$ and $\mathrm{in}(a)\subseteq\{P_{k-1}(q+\varepsilon-1),P_{k-1}(q+\varepsilon)\}$, and $D-1,D,D-\varepsilon,D-\varepsilon+1$ lie in $[1,6(k-1)-1]$; so $a_1\not\sim g$ and $g_1\not\sim a$. Since $a_1,g_1\in S_{k-1}$ are distinct and non-adjacent and $k-1\ge2$, Lemma 11.4(d) gives $g_2\in\mathrm{in}(g_1)\setminus\mathrm{in}(a_1)$; it is not adjacent to $a_1$ (a site of $S_{k-2}$ adjacent to $z\in S_{k-1}$ lies in $\mathrm{in}(z)$) nor to $a$ (layer rule).\qedsq In contrast with $\mathbb Z^d$ there is no corner obstruction: the only sites of $S_k$ adjacent to the inner neighbour $(k-1)\xi_i$ of a corner $k\xi_i$ are the corner and its two neighbours on the cycle $S_k$, which are adjacent to the corner. \textbf{Lemma 11.8} (moving the terminals outward). \textit{Let $\rho\ge3$ and $1\le k\le\rho-2$, let $a\ne g$ be sites of $S_\rho$, let $x,y\in S_k$ be distinct, and let $x'\in\mathrm{out}(x)$, $y'\in\mathrm{out}(y)$ be such that $\{x,x'\}$ and $\{y,y'\}$ do not touch. If $(\mathsf R',\mathsf G')$ is a design for $(a,g;x',y')$ in $B_\rho$ with gadget ball $B_{k+1}$, then adding to each colour the site of $\{x,y\}$ adjacent to its terminal gives a design for $(a,g;x,y)$ with gadget ball $B_k$.} \textit{Proof.} Say $\mathsf R'\cap B_{k+1}=\{x'\}$ and $\mathsf G'\cap B_{k+1}=\{y'\}$, and put $\mathsf R=\mathsf R'\cup\{x\}$, $\mathsf G=\mathsf G'\cup\{y\}$. (D1)–(D3) are clear, using $|x'|=k+1$. For (D4): $x$ and all its neighbours lie in $B_{k+1}$, where $\mathsf G$ consists of $y'$ and $y$, and $\{x,x'\}$, $\{y,y'\}$ do not touch; similarly for $y$.\qedsq \textbf{Lemma 11.9} (routing between two rings). \textit{Let $n\ge1$, $\rho=n+4$, let $T_1,T_2\in S_n$ with $d(T_1,T_2)\ge12$, and let $a,g\in S_\rho$ be distinct and non-adjacent. Then there is a design for $(a,g;T_1,T_2)$ in $B_\rho$ with gadget ball $B_n$.} \textit{Proof.} \textit{Ports.} Lemma 11.7 gives $a_1,g_1,g_2$; choose $g_3\in\mathrm{in}(g_2)$, and let $A=(a,a_1)$ and $B=(g,g_1,g_2,g_3)$. They do not touch ($g_3\in S_{n+1}$ by the layer rule). \textit{Terminals.} For $\nu=1,2$ choose an outward path $\Lambda_\nu=(T_\nu,\lambda^1_\nu,\lambda^2_\nu,\lambda^3_\nu)$, $\lambda^l_\nu\in S_{n+l}$. If both touched $B$, then $d(T_1,T_2)\le3+1+3+1+3=11$; so some $\Lambda=\Lambda_\mu$ does not touch $B$. Let $\mu'$ be the other index and $M=(T_{\mu'},\lambda^1_{\mu'})$; $\Lambda$ and $M$ do not touch, since otherwise $d(T_1,T_2)\le5$. \textit{Upper ring.} $I=N[B]\cap S_{n+3}$ consists of at most five consecutive sites of the cycle $S_{n+3}$ around $g_1$ (Lemma 11.3), and $|S_{n+3}|\ge24$, so $S_{n+3}\setminus I$ is a path. It contains $a_1$ and $\lambda^3_\mu$; let $\Gamma$ be the subpath between them. \textit{Lower ring.} Likewise $I'=N[\Lambda]\cap S_{n+1}$ consists of at most five consecutive sites, and $S_{n+1}\setminus I'$ contains $g_3$ and $\lambda^1_{\mu'}$; let $\Gamma'$ be the subpath between them. \textit{Design.} $\mathsf R=A\cup\Gamma\cup\Lambda$ and $\mathsf G=B\cup\Gamma'\cup M$ satisfy (D1)–(D3), with $\mathsf R\cap B_n=\{T_\mu\}$ and $\mathsf G\cap B_n=\{T_{\mu'}\}$. For (D4): $A$–$B$, $\Lambda$–$B$, $\Lambda$–$M$, $\Gamma$–$B$ and $\Gamma'$–$\Lambda$ were arranged above, and the remaining pairs lie in $S_{n+3}\cup S_{n+4}$ and $S_n\cup S_{n+1}$ respectively.\qedsq \textbf{Theorem 11.10} (TP$_\emptyset$). \textit{Let $m\ge1$ and $\rho\ge m+15$, let $(t_+,t_-)$ be an escapable pair of $S_m(c)$, and let $a,g$ be distinct non-adjacent sites of $S_\rho(c)$. Then there is a design for $(a,g;t_+,t_-)$ in $B_\rho(c)$. Hence TP$_\emptyset$ holds with $\rho_{\rm TP}=16$ and $C_0=15$.} \textit{Proof.} Take $c=o$ and $n=\rho-4\ge m+11$. Let $u_\pm$ be escape witnesses, $x_{m+1}=u_+$, $y_{m+1}=u_-$; they are distinct and non-adjacent. Applying Lemma 11.5 for $k=m+1,\dots,n-1$ gives $x_k,y_k\in S_k$ with $r_k(x_k,y_k)\ge k-m+1$, such that consecutive steps do not touch. Then $r_n(x_n,y_n)\ge12$ and $n\ge12$, so $d(x_n,y_n)\ge12$ by Lemma 11.4(b). Lemma 11.9 gives a design for $(a,g;x_n,y_n)$ with gadget ball $B_n$, and Lemma 11.8, applied for $k=n-1,\dots,m+1$ and finally to $(t_+,t_-,u_+,u_-)$, gives a design for $(a,g;t_+,t_-)$. Spheres of TRI have no pockets, $X=\emptyset$, and $C_0=15\ge\kappa+1$.\qedsq The constant $C_0=15$ is not optimal, but $C_0\ge3$ is necessary: for $\rho=m+2$, the ports $a=P_\rho(i\rho-1)$, $g=P_\rho(i\rho+1)$ and a terminal pair containing $m\xi_i$ admit no design, because both coloured sets must avoid the common inner neighbour $(\rho-1)\xi_i$ of $a$ and $g$, so they contain $P_{\rho-1}(i(\rho-1)\pm1)$, both of which are adjacent to $m\xi_i$. Compared with the outline of [BBR, §6], the terminals here are an arbitrary escapable pair, as produced by Lemma G, rather than two fixed antipodal sites; Lemmas 11.5 and 11.8 move them outward and separate them. \textit{Proof of Theorem B-TRI.} (B1) and (B2) follow from Proposition 11.6 and TP$_\emptyset$ from Theorem 11.10, and (B3) is proved in Section 11.4 (a geodesic from $u$ to $w$ followed by a straight outward segment, separated by one of the functionals $f_i$).\qedsq \subsection*{11.4. The outer boundary property} \textbf{Proposition 11.11.} \textit{(B3) holds for TRI with $c_3=20$, $c_4=30$ and $K_3=1$.} \textit{Proof.} Let $K\ge1$, $L\ge4K+120$ and $L-K-10\le|v|\le L$. Put $W=L-K-20$ and $K'=K+30$. Following inner neighbours from $v$ we reach $w\in S_W$ with $d(v,w)\le K+20$, and $K+30p_c$. Since $p'>p_c$, $\theta(p')>0$. By Lemma B.5(3) and Lemma B.8, \[ \theta(p')\le\tau_L(p',0)\le\tau_L(p',s_0-T)\le\tau_L(p,s_0)\le\tau_L(p,s)\qquad(L\ge L_0), \] and Lemma B.5(2) gives $\theta_E(p,s)\ge\theta(p')>0$.\qedsq The uniformity of $R$ is used through $M=M_R$, and that of $L_0$ in the last display. The hypothesis $p_c<1$ is used to find $p'\in(p_c,p_2]$ inside a compact window on which $g\ge g_0>0$. \subsection*{B.7. Roots} \textit{Finite energy (Remark 6.3).} If $\pi$ is the set of sites of a path from $y$ to $x$, then $\theta_y(p)\ge p^{|\pi|}\theta_x(p)$. Indeed, on $\{\pi\subseteq\omega\}$ the event that $x$ lies in an infinite component of $\omega$ coincides with the event $I$ that $x$ lies in an infinite component of $\omega\cup\pi$, and on both it implies that $y$ lies in the same component. The event $I$ depends only on $\omega\setminus\pi$, so it is independent of $\{\pi\subseteq\omega\}$, and it contains the event that $x$ lies in an infinite component of $\omega$. Hence $\theta_y(p)\ge\mathbb P_p(I)\,p^{|\pi|}\ge p^{|\pi|}\theta_x(p)$. So positivity of $\theta$ does not depend on the root, and neither does $p_c(G)$. The analogous statement for $\theta_E$ can fail on general periodic lattices when the enhancement is not monotone (Remark 6.3): on $\mathbb Z^2$ with a pendant site attached to every site, the non-essential enhancement that opens a site of $\mathbb Z^2$ and its four neighbours in $\mathbb Z^2$ whenever its ball of radius $3$ contains no open site has $\theta_E>0$ at the sites of $\mathbb Z^2$ for small $p>0$ and $s$ close to $1$, while $\theta_E=0$ at the pendant sites for $p<\frac14$. \textit{Remark 6.4.} Let $\phi$ be an automorphism with $\phi\tau_y=\tau_{Ay}\phi$. Then $E_{\tau_yu}(\tau_y\omega)=\tau_yE_u(\omega)$ gives $E^\phi_{\tau_zv}(\tau_z\omega)=\tau_zE^\phi_v(\omega)$, so $E^\phi$ is the enhancement with rules $E^\phi_j=E^\phi_{(0,j)}$, of range $r_E$. If $\phi^{-1}(0,j)=(y,i)$, then $\omega\cup E^\phi_j(\omega)=\phi\tau_y(\tilde\omega\cup E_i(\tilde\omega))$ with $\tilde\omega=\tau_{-y}\phi^{-1}\omega$, so $E^\phi$ is essential iff $E$ is. Finally $E^\phi(\phi\omega,\phi\alpha)=\phi E(\omega,\alpha)$, and $(\phi\omega,\phi\alpha)$ has law $\mathbb P_{p,s}$. For HEX, $\phi(z,i)=(-z,1-i)$ works with $A=-I$. For KAG, the map $a(z)\mapsto b(Mz)$, $b(z)\mapsto c(Mz)$, $c(z)\mapsto a(Mz)$ with $Me_1=-e_1+e_2$, $Me_2=-e_1$ maps the edges $a(z)b(z)$, $b(z)c(z)$, $c(z)a(z)$, $a(z)b(z-e_1)$, $a(z)c(z-e_2)$, $b(z)c(z+e_1-e_2)$ to edges of KAG; it is bijective and maps the four edges at each site injectively, hence onto the four edges at the image, so it is an automorphism, with $A=M$. Its powers permute the three orbits cyclically. \section*{Appendix C. Tables for the periodic lattices} This appendix contains three tables referred to in the text: Table 1, on the exceptional families of the two-port property (Section 1.4), and Tables 5 and 6, on the computations and on the sources in [S] of the proofs for the lattices of Section 12 (Section 12.3, whose convention for statement numbers of [S] applies to them). Table 1 lists the exceptional families of the two-port property (Section 7.2) used on the lattices of Theorem 1.2 and on the four lattices of Theorem 1.5, and the number of port pairs on a large sphere that admit no design (Definition 7.2). For the honeycomb, snub square and snub hexagonal lattices the family consists exactly of these pairs. The families $X^{\rm gen}(2)$ contain them and further pairs. On $\mathbb Z^d$ the exceptional pairs are the corner pairs of Definition 2.11, $4d(d-1)$ on every sphere of radius at least $2$. \begingroup\setlength{\tabcolsep}{4pt} \begin{longtable}{P{103.7pt}P{12.1pt}P{37.2pt}P{102.7pt}P{131.1pt}P{39.4pt}} \toprule \textbf{lattice} & \textbf{$k$} & \textbf{degree} & \textbf{exceptional family} & \textbf{failing non-adjacent pairs per large sphere} & \textbf{section} \\ \midrule \endfirsthead \toprule \textbf{lattice} & \textbf{$k$} & \textbf{degree} & \textbf{exceptional family} & \textbf{failing non-adjacent pairs per large sphere} & \textbf{section} \\ \midrule \endhead honeycomb & $2$ & $3$ & $X^{\rm hex}$ (corner pairs) & $12$ & 9 \\ Kagomé & $3$ & $4$ & none & $0$ & 10 \\ triangular & $1$ & $6$ & none & $0$ & 11 \\ elongated triangular $3^3.4^2$ & $2$ & $5$ & $X^{\rm gen}(2)$ & $4$ & 12.4 \\ snub square $3^2.4.3.4$ & $4$ & $5$ & $X^{\rm snub}$ & $2$ & 12.5 \\ snub hexagonal $3^4.6$ & $6$ & $5$ & $X^{\rm sc}$ & $2$ or $3$ & 12.6 \\ rhombitrihexagonal $3.4.6.4$ & $6$ & $4$ & $X^{\rm gen}(2)$ & $6$ & 12.7 \\ square with diagonals & $1$ & $8$ & $X^{\rm gen}(2)$ & $8$ & 12.8 \\ face-centred cubic & $1$ & $12$ & $X^{\rm gen}(2)$ & $24$ & 12.9 \\ truncated square $4.8^2$ & $4$ & $3$ & $X^{\rm cor}$ and the thin pairs (TP$^*$) & the thin pairs and $4$ (computed) & 12.13 \\ truncated hexagonal $3.12^2$ & $6$ & $3$ & $X^{\rm T}$ (triangle-blocked pairs) & at least $\lfloor(\rho+1)/4\rfloor$ & 12.14 \\ truncated trihexagonal $4.6.12$ & $12$ & $3$ & $X^{\rm cor}$ on $\mathcal R$ and the locally blocked pairs & the locally blocked pairs; the dead and hexagon pairs among them grow linearly & 12.15 \\ diamond & $2$ & $4$ & $D_\rho(c)\cup V_\rho(c)$; corner family $X^{\rm vtx}$ & the dead pairs, growing linearly, and the $24$ vertex pairs & 12.16 \\ \bottomrule \end{longtable} \endgroup \textit{Table 1.} Here $k$ is the number of orbits of sites under the translations of the coordinates used (Definition 6.1). The fifth column gives the number of pairs of distinct non-adjacent non-pocket sites of a large sphere that admit no design (Theorems B-HEX, B-KAG and B-TRI of Sections 9–11, and Theorem 12.2). For the families $X^{\rm gen}(2)$ these are some of the non-adjacent pairs of the family (all of them on the square lattice with diagonals). On the elongated triangular, rhombitrihexagonal and face-centred cubic lattices the other non-adjacent pairs of the family have designs (Section 12.2). For the truncated square lattice the family consists of the four corner pairs of $X^{\rm cor}$ and the thin pairs (two opposite vertices of a square), whose number grows linearly with the radius; the fifth column gives the result of exact computations at the radii $8\le\rho\le17$ and $20\le\rho\le23$ ([S, Part S2, Remark 2.5]), which is not proved for other radii and is not used (Section 12.13). For the truncated hexagonal lattice the port pairs in the family $X^{\rm T}$ are exactly the port pairs without a design when $\rho\ge\max(111,m+28)$ ([S, Part X2, Corollary 8.4]); their number grows linearly with the radius, and the corner lemma CL$_{X^{\rm T}}$ (Section 1.4) holds for another reason (Section 12.14). For the truncated trihexagonal lattice the port pairs without a design are, for gadget radii $m\ge577$, $\rho\ge m+32$, gadget centres $c'$ with $d(c,c')\le3$ and terminals that satisfy the tail condition TF$_2$, exactly the locally blocked pairs ([S, Part T2, Corollary 10.2]); the corner lemma is applied to the subfamily $X^{\rm cor}$ along a set $\mathcal R$ of radii, and the other classes are treated as described in Section 12.15. For the diamond lattice the port pairs without a design are, for $m\ge98$ and $\rho\ge m+17$, exactly the dead pairs and the $24$ vertex pairs; the corner lemma is applied to the subfamily $X^{\rm vtx}$ of the $48$ pairs that contain a vertex site of the sphere, and the edge pairs are treated with history hosts (Section 12.16). \begingroup\setlength{\tabcolsep}{4pt} \begin{longtable}{P{55.7pt}P{219.6pt}P{174.9pt}} \toprule \textbf{lattice} & \textbf{statements whose proofs use a finite computation} & \textbf{extension to all radii} \\ \midrule \endfirsthead \toprule \textbf{lattice} & \textbf{statements whose proofs use a finite computation} & \textbf{extension to all radii} \\ \midrule \endhead KING & none & — \\ ELTRI & apexes ([S, Part E2, Lemma 2.4]), (B1) ([S, Part E2, Proposition 3.1(b)]), tracks ([S, Part E2, Lemma 7.2(b)]), the three local lemmas ([S, Part E2, Lemmas 7.5–7.7]: ports at $\rho\in\{164,165\}$, terminals for $m\le125$, rays for $n\le60$) & local types, [S, Part E2, Lemmas 1.4, 1.5 and Corollary 1.6] \\ SNUBSQ & descent in the distance formula, spheres and apexes for $3\le\rho\le69$, tracks, rays, (B1), the local lemmas of [S, Part Q2, Theorem 8.3], parts of (B3) (certificates C0–C4, C6–C8 of [S, Part Q2]); apexes for $2\le\rho\le7$ and the base case $D\le7$ of the distance formula ([S, Part QC, Theorem 1.6]) & window reduction, [S, Part Q2, Proposition 3.3] \\ SNUBHEX & descent in the distance formula ([S, Part H2, Lemmas 1.0, 1.2]), [S, Part H2, Lemma 2.1], the family $X^{\rm sh}$ ([S, Part H2, Proposition 2.6 and Remark 2.7]), (B1) ([S, Part H2, Proposition 3.1]), tracks ([S, Part H2, Lemma 4.2(c)]), parts of (B3) ([S, Part H2, Lemmas 5.2–5.4]), the local lemmas ([S, Part H2, Lemmas 7.4–7.7]); the descent of [S, Part HC, Lemma 2.6], [S, Part HC, Lemma 5.3] for $2\le\rho\le16$, and (X4) for $2\le\rho\le52$ ([S, Part HC, Proposition 6.5]) & transfer lemma, [S, Part H2, Lemma 2.3]; by hand in [S, Part HC] \\ RTH & the $24$ sites of level $1$ to $3$ ([S, Part R2, Theorem 1.3]), (B1) for $m\le40$, tracks, apexes for $3\le n\le60$, the local lemmas ([S, Part R2, Lemmas 8.6–8.8]: ports at $\rho\in\{195,196,197\}$), a step of (B3) & transfer principle, [S, Part R2, Section 3] (Lemmas 3.1, 3.2, 3.4, 3.5 and Corollary 3.3) \\ FCC & (B1) for $m\le11$ ([S, Part F2, Proposition 3.1]), the local lemmas ([S, Part F2, Lemmas 7.5–7.7]); none in [S, Part FC] & $R$-types, [S, Part F2, Definition 7.1 and Lemmas 7.2, 7.3] \\ BCC & the connectivity of a track minus a spoke for $3\le k\le30$ and a local form of it at $k=7$ ([S, Part B2, Proposition 9.5]), the local port escapes for $13\le\rho\le53$ and terminal escapes for $20\le m\le53$ ([S, Part B2, Proposition 9.7]), the local port escapes in history hosts for $30\le\rho\le59$ ([S, Part B2, Proposition 9.8]); none in [S, Part BC] & connectivity of tracks minus a ball and transfer lemma, [S, Part B2, Lemmas 9.4, 9.6] \\ $4.8^2$ & (X2)–(X4) for $X^{\rm cor}$ ([S, Part S2, Proposition 3.1], exact for all radii), ascent and descent of faces with top $\le250$ (Lemma 6.3), unique outer neighbours for levels $1$–$140$ (Lemma 7.1), the finite inputs of (B3) at levels $491$–$502$ and for $8$ representatives (Lemmas 7.8, 7.9), rays crossing tracks for $10\le n\le250$ (Lemmas 8.1(a), 8.2), the local escapes (Lemmas 8.4, 8.5: ports at $387\le\rho\le398$, terminals for $411\le m\le422$) & exact for the corner data; $R$-types, [S, Part S2, Lemmas 5.1, 5.3 and Corollary 5.2], and translation along the rays (proof of Lemma 8.2) \\ $3.12^2$ & the gadget radii ([S, Part X2, Theorem 4.6]: $m\le64$, $m\not\equiv2\pmod4$), the second bound of Lemma 7.6(c) (levels $\le32$), rays crossing tracks (Lemma 7.7, $2\le n\le40$), the local escapes (Lemmas 8.1, 8.2: ports at $111\le\rho\le115$, terminals for $1\le m\le115$) & transfer principle, [S, Part X2, Lemmas 2.2, 2.3]; translation along the rays (Lemma 7.7); the first bound of Lemma 7.6(c), by hand, for levels $\ge10$ \\ $4.6.12$ & the lattice structure ([S, Part T2, Lemma 1.2(a)–(d)]), the distance formula ([S, Part T2, Theorem 2.1 and Corollary 2.2]), the transfer bounds (Lemma 2.4(b)), the classification of the locally blocked pairs (Proposition 3.4: $8\le r\le356$), the corner data (Proposition 5.2), the descent property of the tracks (Lemma 7.3(a): $n\le250$), rays (Lemmas 8.2, 8.4: $n\le280$), the local escapes (Lemmas 9.1, 9.2, 10.3: ports at $513\le\rho\le522$, terminals for $577\le m\le586$), the local steps of (B3) (Lemmas 11.1–11.3); in the appendix of [S, Part T2]: the absence of pockets ($r\le300$), the thresholds $M^*(c)$ ([S, Part T2, Lemma A.5.4], and $M^*(32)=266$) and the dead-pair test of [S, Part T2, Proposition A.4.6(a)] ($\rho\le300$), used in [S, Part T2, Theorem 4.3] & the lattice structure and $M^*(c)$: finite objects, no extension needed; the distance formula: induction ([S, Part T2, Theorem A.5.3]) and an exact identity of affine functions; the transfer bounds: complete enumeration; the classification, the descent property, the crossings of rays with tracks, the local escapes and the local steps of (B3): transfer lemmas, [S, Part T2, Lemmas 2.5 and 2.6]; the rays: cone lemma and periodicity, and the partner rays of (B3) also through them; the absence of pockets and the dead-pair test: [S, Part T2, Lemma A.5.5]; the corner data exactly for all $\rho\in\mathcal R$ \\ DIA & the bound on the history chains ([S, Part D2, Theorem 5.9]) and the local escapes for close port pairs, close terminal pairs and history hosts ([S, Part D2, Proposition 7.8]), at finitely many radii and class representatives & transfer lemmas, [S, Part D2, Lemmas 1.8 and 7.2] \\ \bottomrule \end{longtable} \endgroup \textit{Table 5.} The certificates (programs, commands, outputs and the second encodings) are described in [S, Part Z] and archived as stated under Data availability. \begingroup\setlength{\tabcolsep}{4pt} \begin{longtable}{P{55.7pt}P{64.0pt}P{76.4pt}P{67.0pt}P{76.1pt}P{87.1pt}} \toprule \textbf{lattice} & \textbf{spheres} & \textbf{(B1); (B2); (B3)} & \textbf{two-port property} & \textbf{corner data} & \textbf{$p_c$} \\ \midrule \endfirsthead \toprule \textbf{lattice} & \textbf{spheres} & \textbf{(B1); (B2); (B3)} & \textbf{two-port property} & \textbf{corner data} & \textbf{$p_c$} \\ \midrule \endhead KING & {}[S, Part K2, Lemma 2.2] (neighbours in adjacent spheres), Lemmas 2.4, 2.5 (outward rays, diagonals) & {}[S, Part K2, Propositions 3.1, 4.1, 5.1] & {}[S, Part K2]: port escapes, Lemmas 7.1, 7.2; terminal escapes, Lemma 7.3; crossing of blocks by paths of increasing level, Lemma 2.6; design, Theorem 7.4 & {}[S, Part KC, Theorem K and Corollary K]; Propositions 2.1, 3.1, 4.1, 5.2 & $1/7\le p_c(\mathrm{KING})\le p_c^{\mathrm{site}}(\mathbb Z^2)<1$, [S, Part K2, Proposition 8.1] \\ ELTRI & {}[S, Part E2, Lemmas 2.1, 2.2] (spheres), Lemma 2.4 (apexes) & {}[S, Part E2, Propositions 3.1, 4.1, 5.4]; (B3) also uses Lemmas 5.1–5.3 & {}[S, Part E2]: tracks, Definition 7.1, Lemma 7.2; apex rays, Definition 7.3, Lemma 7.4; port escapes, terminal escapes, rays crossing tracks, Lemmas 7.5, 7.6, 7.7; design, Theorem 7.9 & {}[S, Part EC, Theorem E]; via the functionals $F_\pi$, Lemmas 2.2, 2.3, Proposition 3.2, Theorem 3.3, Propositions 4.1, 5.1, 6.3 & $\frac14\le p_c(\mathrm{ELTRI})\le p_c^{\mathrm{site}}(\mathbb Z^2)^{1/2}<1$, [S, Part E2, Proposition 8.1] \\ SNUBSQ & {}[S, Part Q2, Lemma 2.1] (no pockets), Lemma 2.3 and Corollary 2.4 (number of apexes) & {}[S, Part Q2, Propositions 5.1, 6.1, Theorem 9.4]; (B3) also uses Lemmas 9.1–9.3 & {}[S, Part Q2]: tracks are induced cycles, Theorem 4.2; corner rays and their crossings with tracks, Lemmas 4.4, 4.5; port and terminal escapes, Lemmas 8.1, 8.2; design, Theorem 8.3 & {}[S, Part QC, Theorem III]; [S, Part QH, Theorem 1, Lemma 1.1, Corollary 2.3] & $\frac14\le p_c(\mathrm{SNUBSQ})\le p_c^{\mathrm{site}}(\mathbb Z^2)^{1/4}<1$, [S, Part Q2, Proposition 10.1] \\ SNUBHEX & {}[S, Part H2, Lemma 2.1] (inner and outer neighbours), Proposition 2.6 (dead pairs) & {}[S, Part H2, Propositions 3.1, 4.3, 5.1]; (B2) uses tracks, Definition 4.1, Lemma 4.2, and (B3) also uses Lemmas 5.2–5.4 & {}[S, Part H2]: apex rays, Definition 7.3, Lemma 7.4; local lemmas, Lemmas 7.5–7.7; design, Theorem 7.9 & {}[S, Part HC, Theorems 5.4, 6.1]; [S, Part QH, Theorem 3.10] & $\frac14\le p_c(\mathrm{SNUBHEX})\le p_c^{\mathrm{site}}(\mathbb Z^2)^{1/6}<1$, [S, Part H2, Proposition 8.1] \\ RTH & {}[S, Part R2, Definition 2.1, Lemma 2.2] (hexagonal spheres with a boundary coordinate) & {}[S, Part R2, Propositions 5.1, 6.1, 7.3]; (B3) also uses Lemmas 7.0–7.2 & {}[S, Part R2]: tracks, Definition 8.1, Theorem 8.2; corner rays, Definition 8.3, Lemma 8.4; local lemmas, Lemmas 8.6–8.8; design, Theorem 9.2 & {}[S, Part RC, Theorem R, Proposition 5.3]; in the cube coordinates, Lemma 2.2, Propositions 2.3, 2.4, 3.1, Lemma 5.1 & $\frac13\le p_c(\mathrm{RTH})\le p_c^{\mathrm{site}}(\mathbb Z^2)^{1/6}<1$, [S, Part R2, Proposition 4.1] \\ FCC & {}[S, Part F2, Lemma 2.2] (spheres), Lemma 2.5 and Corollary 2.6 (a sphere minus a small box around a site is connected) & {}[S, Part F2, Propositions 3.1, 4.1, 5.4]; (B3) also uses Lemmas 5.1–5.3 & {}[S, Part F2]: vertex rays, Lemma 7.4; port and terminal escapes, Lemmas 7.5, 7.6; connectivity of a sphere minus a vertex ray, Lemma 7.7; design, Theorem 7.8 & {}[S, Part FC, Theorem F, Corollary F2, Proposition 5.4]; Lemma 3.2, Lemmas 5.1–5.3 & $1/11\le p_c(\mathrm{FCC})\le p_c^{\mathrm{site}}(\mathbb Z^2)<1$, [S, Part F2, Proposition 8.1] \\ BCC & {}[S, Part B2, Lemmas 1.2 and 1.3] & {}[S, Part B2, Lemmas 5.3, 5.4, 5.5] & {}[S, Part B2]: design, Proposition 9.2; TP$^{\rm bcc}$ and HTP$^{\rm bcc}$, Theorem 9.9 & {}[S, Part BC, Proposition 3.1] (for $X^{\rm cor}$) & $1/7\le p_c(\mathrm{BCC})\le p_c^{\mathrm{site}}(\mathbb Z^2)^{1/2}<1$, [S, Part B2, Remark 5.2] \\ $4.8^2$ & {}[S, Part S2, Lemma 1.2] (distance formula), Corollary 1.3 (inner and outer neighbours, no pockets), Remark 1.6 (spheres along a side) & {}[S, Part S2, Propositions 7.2 ((B1$'$)), 7.3, 7.10]; (B2) uses tracks, Proposition 6.4, and (B3) also uses Lemmas 7.4–7.9 & {}[S, Part S2]: tracks, Definition 6.1, Lemma 6.3, Proposition 6.4; axis rays, Lemmas 8.1, 8.2; port and terminal escapes, Lemmas 8.4, 8.5; design, Theorem 8.6; thin pairs, Lemma 2.2; the reduction, Theorem 4.2 & {}[S, Part S2, Proposition 3.1] (for $X^{\rm cor}$), Corollary 3.2 (deep-occupied corner lemma) & $\frac12\le p_c(4.8^2)\le p_c^{\rm site}(\mathbb Z^2)^{1/4}<1$, [S, Part S2, Proposition 1.5] \\ $3.12^2$ & {}[S, Part X2, Lemmas 1.3, 1.4] (darts, distance formula), Corollaries 1.5, 1.7 (levels, pockets) & {}[S, Part X2, Theorem 4.6 and Lemma 4.5] (gadget radii in place of (B1)); Theorem 5.1; Theorem 6.2 ((B3$_{\rm odd}$)), with Theorem 4.8 (Claim 5 along odd $L$) & {}[S, Part X2]: tracks, Definition 7.2, Lemmas 7.1, 7.3; rays, Definition 7.4, Lemmas 7.6, 7.7; local escapes, Lemmas 8.1, 8.2; design, Theorem 8.3; failing pairs, Proposition 3.5 and Corollary 8.4 & none: CL$_{X^{\rm T}}$ by [S, Part X2, Lemma 4.1 and Corollary 4.2]; no family with (X2)–(X3), Theorem 3.7 & $\tfrac12\le p_c(\mathrm{TH})\le p_c^{\rm site}(\mathbb Z^2)^{1/6}<1$, [S, Part X2, Lemma 1.8] \\ $4.6.12$ & {}[S, Part T2, Theorem 2.1 and Corollary 2.2] (distance formula) & (B1) not used (non-escapable pairs exist for $3\le m\le250$, by computation); the terminals satisfy TF$_2$ ([S, Part T2, Lemma 6.1]); [S, Part T2, Propositions 7.4, 11.4] & {}[S, Part T2]: locally blocked pairs, Definition 3.1, Proposition 3.4; hexagon pairs and their hosts, Lemmas 4.4, 4.6; tracks, Lemma 7.2; TP and HTP, Theorems 10.1, 10.4; the reduction, Theorem 12.1 & {}[S, Part T2, Proposition 5.2] (for $X^{\rm cor}$ on $\mathcal R$), Theorem 6.2 and Corollary 6.3 (corner lemma along $\mathcal R$) & $\frac12\le p_c(\mathrm{TT})\le p_c^{\rm site}(\mathbb Z^2)^{1/12}<1$, [S, Part T2, Lemma 1.3] \\ DIA & {}[S, Part D2, Lemma 1.2] (distance formula), Corollary 1.3 (spheres, no pockets) & {}[S, Part D2, Propositions 3.3, 3.4, 3.9] & {}[S, Part D2]: dead pairs and vertex pairs, Theorem 2.3, Proposition 2.4; three-layer tracks, Lemma 3.1, Corollary 3.2; history hosts, Definition 5.8, Theorem 5.9, Corollary 5.10; assembly, Proposition 7.6; far pairs, Lemma 7.7; TP$^{\rm dia}$ and HTP$^{\rm dia}$, Theorem 7.9; the reduction, Theorem 6.2 & {}[S, Part D2, Proposition 4.2] (for $X^{\rm vtx}$), Corollary 4.3 (deep-occupied corner lemma) & $\tfrac13\le p_c(\mathrm{DIA})\le p_c^{\rm site}(\mathbb Z^2)^{1/2}<1$, [S, Part D2, Lemma 1.5] \\ \bottomrule \end{longtable} \endgroup \textit{Table 6.} The failing pairs are treated in the subsections on the individual lattices, and the finite computations and their extension to all radii are listed in Table 5. For the lattices of Theorem 1.2 the bounds on $p_c$ are those of Proposition 6.6. \section*{Appendix D. Names of the formal statements} The table gives, for the main statements of the paper whose formal versions Section 14 mentions, the names of the corresponding theorems of the Lean formalization, in code font and exactly as in the formalization. Section 14 says how far each formal statement goes. \begingroup\setlength{\tabcolsep}{4pt} \begin{longtable}{P{126.5pt}P{331.7pt}} \toprule \textbf{statement} & \textbf{Lean names} \\ \midrule \endfirsthead \toprule \textbf{statement} & \textbf{Lean names} \\ \midrule \endhead Theorem 1.1 & \texttt{BBR.theorem\_1\_1\_unconditional}; with the bounds $0