\section{Technical supplement: the earlier length-131 escape} \subsection{An earlier length-131 escape construction} \label{sec:construction-131} This retained construction documents one explicit route out of the weighted components. Its upper bound is superseded by \Cref{thm:construction-107}. For a lower word $u$ and a nonempty proper letter set $B$, consider the simultaneous cochain change \[ \Delta_{u,B}=\sum_{a\in B}(\ind_{ua}-\ind_{au}) +J\!\left(\sum_{a\in B}(\ind_{ua}-\ind_{au})\right). \] It preserves RC and the sum on every PCR. It does not by itself guarantee Boolean indicators or acyclicity; both must be verified. \begin{proposition}[Four changes of component]\label{prop:construction-131} Start with the weighted terminal $M_0$ for $\rho=49/16$ (row \texttt{k8\_c094}). Apply the four changes in \Cref{tab:patches} in order. All five endpoints are RC MDS of size 8230, with \[ \boxed{L(M_0),\ldots,L(M_4)=141,135,135,133,131.} \] They occupy five distinct ordinary components and each attains the optimum in its own component. The last component is distinct from all 21 in \Cref{tab:component-optima}. \end{proposition} \begin{proof}[Exact certificates] Direct reconstruction checks all $4^8$ membership bits after each patch, and checks RC, one word per PCR and acyclicity. Integer ranks and explicit longest paths give the upper and attained bounds. One-hit simple cycles of lengths $142,136,136,134,132$ give the matching component lower bounds. Ten cycle separators distinguish all pairs of endpoints; another 21 distinguish the last endpoint from the weighted components. Each separator has at most 15 arcs. The separate checker tests these claims without importing the discovery search. \end{proof} \begin{table}[ht] \centering \begin{tabular}{clcrr} \toprule Step & $u$ & $B$ & Changed words & $L$ after\\ \midrule 1&\texttt{ATCGGCG}&$\{T\}$&4&135\\ 2&\texttt{ATTACTG}&$\{G,T\}$&8&135\\ 3&\texttt{CCGCCCG}&$\{C\}$&4&133\\ 4&\texttt{GTTCTTG}&$\{T\}$&4&131\\ \bottomrule \end{tabular} \caption{Explicit simultaneous changes; the neutral second step changes component. Its necessity or minimal number of moves is not claimed.} \label{tab:patches} \end{table} The four changed supports are disjoint. A compact final membership rule is \[ \ind_{M_4}(w)=m_{49/16}(w)-\ind_D(w)+\ind_A(w), \] where the following columns list $D$ and $A$; rows are not canonical pairings. \begin{center} \begin{tabular}{ll} \toprule Remove ($D$)&Add ($A$)\\\midrule \texttt{CAAGAACA}&\texttt{ACAAGAAC}\\ \texttt{CAGTAATA}&\texttt{ACAGTAAT}\\ \texttt{CAGTAATC}&\texttt{ACGCCGAT}\\ \texttt{CCCGCCCG}&\texttt{ATCGGCGT}\\ \texttt{CGCCGATA}&\texttt{ATTACTGG}\\ \texttt{CGGGCGGG}&\texttt{ATTACTGT}\\ \texttt{GATTACTG}&\texttt{CCAGTAAT}\\ \texttt{TATCGGCG}&\texttt{CCGCCCGC}\\ \texttt{TATTACTG}&\texttt{GCGGGCGG}\\ \texttt{TGTTCTTG}&\texttt{GTTCTTGT}\\ \bottomrule \end{tabular} \end{center} \begin{corollary}[Global upper bound, exact component optimum] \label{cor:global-upper-131} For RC DNA MDS at $k=8$, \[ \min_{M\ {\rm RC\ MDS}}L(M)\le131,\qquad \min_{N\in\CF(M_4)}L(N)=131. \] The final upper residual has length 130 in edges. Its longest avoiding word has 138 letters, so every DNA word of length 139 contains a selected eight-letter window. \end{corollary} The final one-hit cycle is the cyclic concatenation of these lines: \begin{center}\ttfamily\small CGCGGAGGAGCCGACCCCACCACACAACAAAAAAACAACACAAGAAGACC\\ AGCAGCCGAGGAGGATAGGCCTATCCTCCTCGGCTGCTGGTCTTCTTGTG\\ TTGTTTTTTTGTTCTTGGGTGCTGGGCTGATG \end{center} Its length is 132, its seven-letter prefixes are distinct, and its only selected cyclic eight-letter window is \texttt{GCGCGGAG} at zero-based index 131. Deleting that arc leaves 131 residual arcs. The machine-readable string in \pathfile{cycle_132_compact.json} is checked against the cycle certificate. Any improvement to $L\le130$ must hit this cycle at least twice, and therefore leave this component. \subsubsection{Reproducibility and evidence boundary} The public package contains the original 30-row regression capsule, the complete 199-row atlas, mixed-wall and component certificates, and the retiming construction. The atlas checks 11,432,704 word decisions. The retiming checker verifies 72 certificate occurrences (not 72 distinct sets), 42 retimings, 30 lower-bound cycles and 31 short separators. Its 24 semantic mutations include missing lower bounds and a duplicated parity sector. The atlas and wall checkers have 18 and 15 further mutation controls respectively. Small exhaustive tests cover $(q,k)=(2,2),(2,4),(4,2)$ and 64 RC-feasibility queries, plus 260 bounded autodual ranks. All finite outputs have a declared scope. A rank certificate proves a universal path bound for its finite DAG; a completed list proves a complete atlas only because the affine-root reduction is exhaustive. None of these checks proves an all-order theorem by sampling. Distinct implementations remain AI-assisted checks, not independent human acceptance. No timeout or incomplete search is used to claim that a shorter global construction is impossible.