\section{Meridians, addresses, and the alternative tree}\label{sec:indexing} \subsection{The landscape and its two coordinates} On the landscape $(\log_{10}M,c)$, the point $(M,c)$ represents the contacts $f_M(c),f_M(c-1)$. We draw the reflected half $00$, $k_i\ne0$ for $i>1$, and $x\in\Z$. Evaluation is followed by sorting the two companions. \end{definition} \begin{theorem}[Configuration bijection]\label{thm:bijection} Canonical marks are in bijection with all normalized positive Markov triples except the root. The displayed value $M_\alpha(x)$ is the largest entry. \end{theorem} \begin{proof} The quadratic argument in Proposition~\ref{prop:closure} shows that replacing the largest coordinate by its other pole gives a smaller maximum, apart from the initial repeated entries. This is the forced Vieta reduction of a configuration and does not assume scalar unicity. To recover a word, reflect the current maximum $H$ to $P=3ab-H$. When $P\ge2$, walk its oriented endpoint orbit inward until both endpoints are below $P$. This edge is unique. Indeed $u_{j+1}-2u_j+u_{j-1}=(3P-2)u_j>0$, so the entries below $P$ form an interval; three consecutive entries below $P$ would contradict $u_{j-1}+u_{j+1}=3Pu_j\ge3P$. Endpoint descent supplies the edge, and its nonzero signed displacement recovers the last-applied, leftmost index. Repeat at the smaller pole. At pole one, the orbit reaches $(1,1)$ and fixes the seed phase. Conversely, all outward blocks are positive and reduced. If the preceding pole is $P6H$ and $(9a^2-2)H-P>6H$. Their farther endpoints exceed $2H$, and the recurrence makes subsequent outward entries increase. The reduction therefore undoes every prescribed block. Finally, companion exchange negates all indices and $x$ simultaneously. The canonical conditions select one of the two orientations; only $(1,1,2)$ is fixed. \end{proof} This bijection is useful computationally as well as geometrically. A configuration can be retained by its address and reconstructed when a comparison needs its full integer entries. Section~\ref{sec:verification} uses the equivalent reduced-fraction labels as compact identities, with compressed Vieta evaluation. Its coverage therefore counts triples before testing whether their maxima are distinct. \subsection{The two parent rules} At a nonzero phase the alternative parent is the adjacent inward mark, $(\alpha;x)\mapsto(\alpha;x-\sgn x)$. At a nonseed helm it is \begin{equation}\label{eq:altparent} (1,k_m,\ldots,k_2,k_1;0)\longmapsto (1,k_m,\ldots,k_2;-k_1), \end{equation} followed by simultaneous sign reflection if the remaining rightmost index is negative. An empty tail uses the seed's nonnegative phase. The seed helm with maximum two has parent $(1,1,1)$. The alternative parent deletes the \emph{rightmost} index at a helm. Recursive evaluation removes the \emph{leftmost} index after $1$. These operations serve different purposes. The depth \begin{equation}\label{eq:depth} d(\alpha;x)=1+\sum_{i=1}^m(|k_i|+1)+|x| \end{equation} decreases by one under either parent edge, with root depth zero. Thus the parent rule defines a rooted tree on the same configurations. The ordinary outward Vieta moves instead are \begin{equation}\label{eq:vietachildren} (a,b,M)\longmapsto(a,M,3aM-b),\qquad (a,b,M)\longmapsto(b,M,3bM-a). \end{equation} For example, $29\to194$ is a Meridian edge, whereas $29\to169$ is a Vieta edge. Figure~\ref{fig:fareytrees} compares the two sets of links. \paperfigure{fareytrees}{.96}{figures/fig2B_farey_trees.pdf}{Two trees on the same 20 configurations: the 16 with $M\le3000$ and the four additional values $6466,7561,14701,37666$. Each node displays its reduced standard Farey label and maximum. Red and green denote the Vieta moves; blue and gold denote Meridian and phasor steps.}{p} \paperfigure{fan}{1}{figures/fig2C_reflection_fan.pdf}{The orbit underlying one recursive extension. Reflecting $H$ across $(u_j,u_{j+1})$ gives $N_j=3u_ju_{j+1}-H$; $N_0$ is the preceding pole. For source $(a,b,H)=(1,5,13)$ at $M_{1,1}(0)$, the preceding pole is two, $(u_{-2},u_{-1},u_2,u_3,u_4)=(1325,34,194,7561,294685)$, and $(N_{-2},N_{-1},N_1,N_2,N_3)=(135137,89,2897,4400489,6684339842)$. The right panel locates these reflections in the Vieta tree.}{htbp} \subsection{Real components and phasor edges} \begin{proposition}\label{prop:realcomponents} The canonical real families in \eqref{eq:word}, modulo companion reflection, are the nonterminal connected components of $H(M,c)=1$. Their terminal phase is $x+1/2$ modulo one. The separate terminal component is $M=1$. \end{proposition} \begin{proof} The seed companions are at least $2/\sqrt5$, and $M_1\ge2$ is strictly larger than either. More generally, let $h\ge1$ and $2/\sqrt5\le a,b\le h$. The first outward endpoint satisfies $3hb-a\ge(6/\sqrt5-1)h>h$; its reflected pole is larger still. The reversed orbit gives the same conclusion on the other side. Every outward block thus has a strict inverse descent to its source and preserves the terminal phase. Conversely, reverse any descent ending at host one. Its last host change produces the seed with $x=v-1/2$, and every preceding walk is nonzero: a zero walk would simply have stopped at the smaller host. These strict blocks uniquely recover a locally constant word. The unrolled $x$ joins the final phase charts continuously. The seed is injective as an ordered curve, as its endpoint difference is strictly monotone; each block is invertible. The positive Laurent expansion proved in Proposition~\ref{prop:laurent} makes every family proper. The inverse charts are relatively open in the level set, and properness makes the families closed. Each connected family therefore fills a component. Reflection gives exactly the canonical identifications. \end{proof} The phasor edges are obtained by applying the same outward blocks to \begin{equation}\label{eq:basearc} \left(\frac h{\sqrt{3h-2}},\frac h{\sqrt{3h-2}},h\right),\qquad 1\le h\le2. \end{equation} This symmetric arc joins $(1,1,1)$ to $(1,1,2)$. Its other host is $2h/(3h-2)$, and both possible terminal fits have phase $1/2$. Its endpoints satisfy the bounds used in the proposition, so the outward blocks preserve that phase. The displayed phasors consequently arise from the descent itself, rather than interpolation between integer nodes. \subsection{Degree and the exact Farey dictionary}\label{subsec:dictionary} Attach endpoint count vectors $u=(1,0)$ and $v=(0,1)$ to the seed. Under an index $k$ update them by \begin{equation}\label{eq:countupdate} (u,v)\longmapsto \begin{cases} ((k-1)(u+v)+v,\ k(u+v)+v),&k>0,\\ (|k|(u+v)+u,\ (|k|-1)(u+v)+u),&k<0. \end{cases} \end{equation} Write their sum as $(r,s)$. Their determinant remains $\pm1$ and their unordered residues modulo two remain $\{(1,0),(0,1)\}$. Thus $r,s$ are positive coprime odd integers. Canonical orientation gives $r\le s$. The \emph{Meridian degree} is $D=r+s$. It is the extreme Laurent exponent in $\rho^x$, where $\rho=(3+\sqrt5)/2$; the degree in $\rho^{2x}$ is $D/2$. \begin{proposition}[Helm classification]\label{prop:helm} Canonical Meridians correspond bijectively to coprime positive odd pairs $r\le s$. Their helm labels are the standard Farey fractions $r/s$. \end{proposition} \begin{proof} The endpoint vectors are the two Stern--Brocot parents: their sum is $(r,s)$ and their determinant has absolute value one. Conversely, the parents of an odd/odd pair have opposite coordinate parities. Their componentwise absolute difference is a smaller positive primitive odd/odd pair, except at $(1,1)$. The difference has one componentwise sign; an opposite sign would give the initial pair. Iteration reaches $(1,1)$. To recover the signed indices, put $n_\alpha=(r,s)$ and let $\alpha^-$ remove the leftmost index after $1$. Consecutive difference vectors have determinant of absolute value two, so three consecutive vectors satisfy $n=2c\,n^-+\varepsilon n^{--}$ with $c\ge1$ and $\varepsilon=\pm1$. Determinant division gives an integer coefficient, parity makes it even, and decreasing coordinate sums fix its magnitude and sign. At the bottom use $(1,1),(-1,1)$; canonical orientation fixes the first index positive. The magnitude is fixed by the ratio of successive degrees, which lies in $[2c,2c+1)$ or $(2c-1,2c]$ according to the sign; equality occurs only at the first applied index. Reversing the process gives precisely $n_{1,k,k_m,\ldots,k_1}=2|k|n_\alpha+\sgn(kk_m)n_{\alpha^-}$. For the seed use $k_0=1$ and $n_{\varnothing}=(-1,1)$, so the first extension has $(r,s)=(2k_1-1,2k_1+1)$. This recovers the unique address. Mediant addition identifies the label. \end{proof} \begin{lemma}[Phase dictionary]\label{lem:dictionary} For helm $(r,s)$ and degree $D=r+s$, the standard Farey label at phase $x$ is \[ \begin{array}{c|c|c} x&(p,q)&p+q\\\hline 0&(r,s)&D\\ 1&(r,D)&D+r\\ -1&(s,D)&D+s\\ n\ge2&(D,(n-1)D+r)&nD+r\\ -n\le-2&(D,(n-1)D+s)&nD+s. \end{array} \] The negative seed phases are omitted. \end{lemma} \begin{proof} At seed phases $0,1,n\ge2$, the endpoint bases are respectively $((1,0),(0,1))$, $((1,1),(0,1))$, and $((1,n),(1,n-1))$. The count updates are linear, so substitution gives the table; negative phase exchanges the bases. For the inverse, odd/odd $p,q$ give $(r,s,x)=(p,q,0)$. If $q$ is even, sort $p,q-p$ and choose $+1$ when $p$ is the smaller entry, otherwise $-1$. If $p$ is even, write $q=tp+u$ with $0