\section{Retiming and reverse-complement sectors} \subsection{Integer retiming and reverse-complement components} \label{sec:retiming} The ordinary statements in this section assume $q\ge2$ and $k\ge2$; the RC statements have the additional hypotheses specified below. This section works on $G=B(q,k-1)$, with lower vertices and length-$k$ arcs, including loops. Write $m=\ind_M$ for an ordinary MDS. A complete move at a residual source $u$ exchanges all incoming and outgoing arcs; its indicator change is $\nabla\ind_u$. Inverse moves act at residual sinks. Let $\CF(M)$ denote the component generated by these moves, allowing non-RC intermediate MDS. \begin{lemma}[Classical component interface]\label{lem:move-component} Two MDS $M,N$ belong to the same ordinary component if and only if \[ \ind_N=m+\nabla f\quad\text{for an integer function }f. \] Thus $m(Q)=\sum_{e\in Q}m(e)$ is constant on each directed cycle throughout the component. \end{lemma} \begin{proof} A move adds a unit gradient, and gradients telescope on cycles. Conversely normalize $\min f=0$. If its maximum $h$ is positive, choose a source $u$ of the current residual induced on $\{f=h\}$. An incoming residual arc from a lower level would give $\ind_N(v\to u)=f(v)-h<0$. Hence $u$ is a source of the entire residual and a complete move is legal. A new cycle is impossible: after the exchange no residual arc leaves $u$. For each letter, $au$ and $ua$ lie in the same PCR, so the one-per-PCR condition makes the exchange Boolean and preserves cardinality. A homopolymer loop stays selected and has zero gradient. Replace $f$ by $f-\ind_u$. Its sum decreases and stays nonnegative, so iteration reaches $N$. \end{proof} The complete-move and cycle-invariant framework is due to Mar\c cais, DeBlasio and Kingsford \cite[Sections 4--5]{MarcaisDeBlasioKingsford2024}. The proof above fixes the gradient sign and the constructive scheduler. \begin{lemma}[Short separation]\label{lem:short-separation} Different ordinary components are separated by a simple directed cycle of length at most $2k-1$ whose selected-arc counts differ. \end{lemma} \begin{proof} Let $d=\ind_N-\ind_M$ and root the graph at $o=a^{k-1}$. Choose $A_u:o\to u$ by appending $u$ and $B_u:u\to o$ by appending $a$; both have length at most $k-1$. Put $p(u)=\sum_{A_u}d$. If some closed walk $A_uB_u$ has nonzero sum, its length is at most $2k-2$. Otherwise $\sum_{B_u}d=-p(u)$, and $A_ueB_v$ has sum $p(u)+d(e)-p(v)$ for $e:u\to v$. If all these sums vanished, $d=\nabla(-p)$, contradicting \Cref{lem:move-component}. A nonzero closed walk of length at most $2k-1$ therefore exists. Decomposing it into simple cycles leaves at least one with nonzero sum and no greater length. \end{proof} \subsubsection{The cycle ratio determines the ordinary optimum} Define \[ R(M)=\max_{Q\text{ simple directed cycle}}\frac{|Q|}{m(Q)}. \] Every denominator is positive because $M$ meets every cycle. \begin{theorem}[Unit retiming]\label{thm:unit-retiming} For every integer $T\ge1$, the following are equivalent: \begin{enumerate} \item Some $N\in\CF(M)$ has $L(N)\le T-1$. \item There is an integer function $H$ satisfying \begin{equation} H(v)-H(u)\ge1-Tm(e)\qquad(e:u\to v). \label{eq:clock} \end{equation} \item Every simple directed cycle $Q$ satisfies $|Q|\le Tm(Q)$. \end{enumerate} Consequently \[ \boxed{\min_{N\in\CF(M)}L(N)=\ceil{R(M)}-1.} \] \end{theorem} \begin{proof} Given $N$ and $\ind_N=m+\nabla f$, let $h$ be its longest-path rank, with values between zero and $T-1$. Then $H=h-Tf$ satisfies \eqref{eq:clock}, on both selected and residual arcs. Conversely divide any integer solution as $H=Ta+h$, with $0\le hi([\theta v])$, otherwise down, leaving integers fixed. Mate roundings are opposite. A tight edge cannot round its tail up and head down: that would require \[ i([\theta u])R(M)$ every cycle has strictly negative clock weight, so the tight graph is acyclic. Its components are singletons, none containing a mate pair. Apply \Cref{thm:rc-rounding} at $T=T_0$ when noncritical, otherwise at $T=T_0+1$. The ordinary optimum supplies the lower bound. \end{proof} \Cref{prop:sharp-parity} supplies an example attaining the extra unit in one prescribed sector. \subsubsection{Exactly two paired-move components} A paired RC step is a complete move at $u$ and its inverse at $\theta u$, with disjoint incident supports. RC is required after the compound step, not after its first half. \begin{theorem}[Exactly two RC sectors]\label{thm:two-rc-sectors} Every ordinary component containing an RC MDS has exactly two components under paired RC steps. They are the two parity sectors. In one sector, if $\ind_N-\ind_M=\nabla g$ and $g\circ\theta=-g$, the paired-step distance is \[ d_{\rm pair}(M,N)=\tfrac12\sum_v|g(v)|. \] Between different sectors the membership of every RC-fixed length-$k$ word is reversed. \end{theorem} \begin{proof} Both sectors exist by \Cref{cor:parity-cost}. In a single sector the difference potential has a unique integer antisymmetric centering $g$. If its maximum is $h>0$, choose a residual source $u$ on that level as in \Cref{lem:move-component}. Its mate is a sink at level $-h$. An edge between them would have gradient $\pm2h$, impossible as the difference of two Boolean indicators. Their stars are disjoint and the paired move is legal. It replaces $g$ by $g-\ind_u+\ind_{\theta u}$, reducing its norm by two. Iteration gives the asserted length; every paired step changes the norm towards the target by at most two, proving optimality. Paired steps preserve parity. For a fixed arc $e:u\to\theta u$, \[ \ind_N(e)-m(e)=2f(u)-\epsilon. \] It is zero for $\epsilon=0$ and $\pm1$ for $\epsilon=1$. Such words exist at even order, so the sectors are distinct. \end{proof} Within the weight-selector family a fixed periodic word $(a\bar a)^{k/2}$ always has the same frozen zero-PCR membership. Thus two terminals in the same ordinary component lie in the same paired sector. The second sector need not be realized by weights. \begin{proposition}[Components encountered by the DNA family] \label{prop:family-components} At $k=4,6,8$ the fixed labelled DNA family $\mathcal W$ with fixed $A1$ in $\mathcal W$. Cut-potential normal forms classify its 199 strata; literal paired moves join each to its assigned representative. The replay checks 3857 compound steps. Every pair of distinct classes has a separating cycle: three at $k=6$ and 210 at $k=8$, all of length at most $2k-1$. These are positive and negative certificates, not an assumption that a search found all paths. The component index is $2\#\{r<\rho\}+\ind_{\rho\text{ is a listed root}}$. A fixed linear form can have only one normalized root; each listed wall is indispensable and one lambda form per root suffices. The periodic anchor proves that the same classification holds for paired moves inside the family. \end{proof} This classifies components encountered by the selector, not all components in the graph. As a small independent boundary check, binary order four has six RC MDS in three ordinary components; all six are isolated under paired moves. Parity cannot be discarded.