\section{Technical supplement: arithmetic proofs and exact counts} \subsection{Component fields, structural zeros, and algebraic designs} \label{sec:components} A zero of a selector form is called \emph{structural} when it holds componentwise, and therefore for every choice of positive magnitudes. A zero that is not structural is called \emph{accidental}. This terminology refers to the weight parameter, not to numerical error. \begin{theorem}[Component-field zero criterion and structural extremality] \label{thm:component-field} Fix an even $k\ge4$. \begin{enumerate} \item If $\beta_1,\ldots,\beta_m$ are linearly independent over $K_k$, then \[ P_s(w)=0 \quad\Longleftrightarrow\quad X_i(w)=0\ \text{for every }i. \] \item If $\beta_1,\ldots,\beta_m$ are linearly independent over $K_k^+$, then \[ \lambda_s(u)=0 \quad\Longleftrightarrow\quad L_i(u)=0\ \text{for every }i. \] \item For arbitrary admissible magnitudes, every componentwise zero remains zero. Hence a field-generic table attains the universal structural floor for each of these two zero sectors. \end{enumerate} \end{theorem} \begin{proof} The decompositions~\eqref{eq:component-decomp} are linear relations in the declared coefficient fields. Under the corresponding field-linear independence, the relation can vanish only when every coefficient vanishes. The converse is immediate. Componentwise vanishing does not involve the $\beta_i$, so those states remain zero for every admissible table. Every other zero is an additional cancellation, proving the extremal assertion. \end{proof} The lower statement would be false as written without dividing out the common factor $\sin(2\pi/k)$. For example, at $k=6$ that sine is $\sqrt3/2$; treating it as a rational coefficient confuses the coefficient field with a common nonzero scalar. \begin{corollary}[DNA field gate] \label{cor:dna-field-gate} For DNA magnitudes $(\rho,1)$, an accidental embedding zero forces $\rho\in K_k$, and an accidental lower zero forces $\rho\in K_k^+$. Field membership is necessary, not sufficient, for a realized DNA cancellation. \end{corollary} \begin{proof} An embedding relation has the form $\rho X_H+X_L=0$. If it is accidental, $X_H\ne0$ and $\rho=-X_L/X_H\in K_k$. The lower proof is identical with $L_H,L_L\in K_k^+$. Membership alone supplies no labelled word whose two component coefficients realize the required quotient. \end{proof} The gate gives exact schedules for probing or avoiding algebraic resonance. We record three useful regimes. All field-theoretic implications below are standard consequences of conductor and ramification theory for cyclotomic fields~\cite{Washington1997}. \begin{corollary}[Three exact weight-design regimes] \label{cor:weight-designs} Let $\rho>1$ be the high DNA magnitude. \begin{enumerate} \item Put $\varphi=(1+\sqrt5)/2$ and $\rho_{\mathrm{gold}}=\varphi^4=(7+3\sqrt5)/2$. Then $\rho_{\mathrm{gold}}\in K_k$ exactly when $5\mid k$. Thus the family is a conductor-five resonance probe. \item If $\rho$ generates a non-Galois real extension of $\Q$, then $\rho\notin K_k$ for every $k$. In particular $\rho=\sqrt[3]{2}$ gives a uniform all-order structural-only design for the two sectors of \Cref{thm:component-field}. \item Given a finite horizon $K$, choose a prime $\ell>K$ with $\ell\equiv1\pmod4$ and put $\rho=\sqrt\ell$. Then $\rho\notin K_k$ for every $4\le k\le K$, so the same two sectors are structural-only throughout the declared horizon. \end{enumerate} \end{corollary} \begin{proof} The quadratic field $\Q(\sqrt5)$ has conductor $5$, which gives the first equivalence. Every subfield of the abelian extension $K_k/\Q$ is Galois over $\Q$; a non-Galois real field therefore cannot be contained in $K_k$. For the last statement, the prime $\ell$ ramifies in $\Q(\sqrt\ell)$, whereas all finite ramification in $K_k$ is supported on primes dividing $k$. \end{proof} These regimes do not assert that the golden family resonates whenever $5\mid k$. They separate where resonance is algebraically impossible from where a certificate-level realization must still be proved. \begin{corollary}[Uniform radical magnitude family] \label{cor:uniform-radical} Fix $m\ge1$, choose an odd prime $d\ge\max\{m,3\}$, put $\alpha=2^{1/d}$, and define \begin{equation} \beta_i=\alpha^{i-1},\qquad 1\le i\le m. \label{eq:uniform-radical} \end{equation} Then the magnitudes are positive and distinct, and they are linearly independent over $K_k$ for every $k$. Consequently, at every even $k\ge4$, the embedding and lower-lambda zero sectors contain only their componentwise structural zeros. \end{corollary} \begin{proof} The polynomial $x^d-2$ is Eisenstein at $2$, so $E=\Q(\alpha)$ has prime degree $d$. For any cyclotomic field $K_k$, the intersection $E\cap K_k$ therefore has degree either $1$ or $d$ over $\Q$. The second case would embed $E$ as an intermediate field of the abelian Galois extension $K_k/\Q$, forcing $E/\Q$ to be Galois. This is impossible: $E\subset\R$, whereas the irreducible polynomial $x^d-2$ has nonreal roots $\alpha\zeta_d^j$ and does not split in $E$. Thus $E\cap K_k=\Q$. Since $K_k/\Q$ is Galois, the two fields are linearly disjoint, and the degree of $\alpha$ over $K_k$ remains $d$. No nonzero polynomial over $K_k$ of degree below $d$ can vanish at $\alpha$; because $m\le d$, the powers $1,\alpha,\ldots,\alpha^{m-1}$ are independent over $K_k$. The conclusion follows from \Cref{thm:component-field}, with independence over $K_k$ also implying independence over $K_k^+$. \end{proof} This family is an exact existence design. It makes no claim about numerical conditioning, magnitude range, downstream path quality, or practical superiority to a DNA weight table. \subsection{The twice-prime structural formula} \label{sec:twice-prime-structural} Let $k=2p$ with $p$ an odd prime and put $h=(p-1)/2$. Identify the letters with the vertices \[ V_m=\{\pm e_1,\ldots,\pm e_m\}\subset\Z^m \] through the component map $x\mapsto(\varepsilon_1(x),\ldots, \varepsilon_m(x))$. \begin{lemma}[Prime sine basis] \label{lem:prime-sines} The positive numbers \[ \sin(\pi/p),\ldots,\sin(h\pi/p) \] are linearly independent over $\Q$. \end{lemma} \begin{proof} Suppose $\sum_{r=1}^h c_r\sin(r\pi/p)=0$ with $c_r\in\Q$, clear denominators, and write $\xi=e^{\pi i/p}$. Then \[ g(x)=\sum_{r=1}^h c_r(x^{h+r}-x^{h-r}) \] satisfies $g(\xi)=0$ and has degree at most $p-1$. The minimal polynomial $\Phi_{2p}(x)=x^{p-1}-x^{p-2}+\cdots-x+1$ also has degree $p-1$. Therefore $g$ is zero or a scalar multiple of $\Phi_{2p}$. The former polynomial is anti-palindromic and the latter palindromic, so comparison of leading and constant coefficients forces the scalar to vanish. The disjoint coefficient pairs of $g$ then give $c_r=0$ for every $r$. \end{proof} For a lower word of length $2p-1$, sine indices $1,\ldots,2p-1$ split into the free central index $p$ and, for $1\le r\le h$, the four-position block \[ \{r,p-r,p+r,2p-r\}. \] The coefficient of $\sin(r\pi/p)$ is therefore \begin{equation} v(u_r)+v(u_{p-r})-v(u_{p+r})-v(u_{2p-r})\in\Z^m. \label{eq:block-vector} \end{equation} \begin{lemma}[Cross-polytope additive energy] \label{lem:cross-energy} \leavevmode\par\noindent The number of ordered quadruples $(x_1,x_2,x_3,x_4)\in V_m^4$ satisfying $x_1+x_2=x_3+x_4$ is \begin{equation} E(V_m)=6m(2m-1). \label{eq:cross-energy} \end{equation} \end{lemma} \begin{proof} Count ordered pairs by their sum. The zero sum has multiplicity $2m$. Each sum $\pm2e_i$ has multiplicity one, contributing $2m$. For each unordered pair of distinct coordinate axes and each independent choice of signs, a sum $\pm e_i\pm e_j$ has multiplicity two. There are $4\binom m2$ such sums. Thus \[ E(V_m)=(2m)^2+2m+4\binom m2\,2^2 =6m(2m-1). \] \end{proof} \begin{theorem}[Twice-prime structural zeros and firings] \label{thm:twice-prime-structural} Let $q=2m$, $k=2p$, $p$ an odd prime, and $h=(p-1)/2$. The number of structural lower zeros is \begin{equation} Z_{\mathrm{str}}(q,2p)=q\,[6m(2m-1)]^h. \label{eq:zstruct} \end{equation} For a table whose magnitudes are linearly independent over $K_{2p}^+$, this is the exact lower-zero count. The direct descent then has \begin{equation} F_{\mathrm{lower}}(q,2p) =\frac{q^{2p-1}-Z_{\mathrm{str}}(q,2p)}4 \label{eq:firing-count} \end{equation} physical lower firings. \end{theorem} \begin{proof} By \Cref{lem:prime-sines}, componentwise lower vanishing is equivalent to the independent vanishing of~\eqref{eq:block-vector} in every one of the $h$ blocks. Each block has $E(V_m)=6m(2m-1)$ ordered assignments by \Cref{lem:cross-energy}; the central letter has sine coefficient zero and is free, giving the factor $q$. This proves~\eqref{eq:zstruct}. Field genericity turns structural into total lower vanishing by \Cref{thm:component-field}. Letterwise complementation changes the sign of every lower value. Thus the nonzero states split equally between the two signs. Reverse complement acts freely on the positive states and the self-dual ideal chooses exactly one state from each pair. Hence the firing count is one quarter of the nonzero lower-state count, proving~\eqref{eq:firing-count}. \end{proof} The factor $q$ in~\eqref{eq:zstruct} is load-bearing: it is the free central coordinate. The count is for ordered letter assignments, and the firing denominator is four rather than two. \begin{corollary}[Closed structural-zero density] \label{cor:structural-zero-density} Under the hypotheses of \Cref{thm:twice-prime-structural}, \begin{equation} Z_{\mathrm{str}}(q,2p)=q\,[3q(q-1)]^h, \qquad \frac{Z_{\mathrm{str}}(q,2p)}{q^{2p-1}} =\left(\frac{3(q-1)}{q^3}\right)^h . \label{eq:structural-density} \end{equation} \end{corollary} \begin{proof} Substitute $q=2m$ into $6m(2m-1)=3q(q-1)$. Since $p=2h+1$, the lower-word length is $2p-1=4h+1$; division by $q^{4h+1}$ gives the second identity. \end{proof} \begin{theorem}[Single-odd-prime structural embedding words and PCRs] \label{thm:twice-prime-embedding} Let $q=2m$, let $p$ be an odd prime, and let \[ k=2^b p^a,\qquad r=2^{b-1}p^{a-1},\qquad a,b\ge1. \] Put \[ M_p(q)=q^p+q+2^{p-1}q(q-2). \] The number of componentwise structural embedding-zero words is \begin{equation} Z_{\mathrm{emb,str}}(q,k)=M_p(q)^r. \label{eq:embedding-word-count} \end{equation} The number of their rotation orbits is \begin{equation} N_{\mathrm{emb,str}}(q,k) =\frac1k\left( M_p(q)^r+ \sum_{\substack{d\mid k\\d0}$. Then, after multiplication by a sufficient monomial to remove negative exponents, \begin{equation} \PhiK(z)\mid z^N\bigl(U(z)A(z)+dB(z)\bigr)\quad\text{in }\Z[z]. \label{eq:certificate-divisibility} \end{equation} Conversely, this divisibility implies the scalar equality for the same retained certificate. \end{enumerate} \end{theorem} \begin{proof} Substitution $s(w_j)=\rho h_j+\ell_j$ into the role's linear equality gives \eqref{eq:certificate-equality} term by term, proving the first assertion. The next two assertions are the exhaustive alternatives for the two evaluated coefficients. When $A(\zetaK)\ne0$, solving a linear equation in $\rho$ gives the unique candidate. For the last assertion, multiply $\rho A(\zetaK)+B(\zetaK)=0$ by $d$ and substitute the represented value of $\rho$. Clearing Laurent exponents gives an integer polynomial vanishing at the primitive root $\zetaK$. Its minimal polynomial over $\Q$ is $\PhiK$, so divisibility follows. Conversely, divisibility implies vanishing at $\zetaK$ and hence the original scalar equality. Because the certificate is retained, the converse returns to the same labelled word and equality role; it does not turn an arbitrary divisible polynomial into a DNA witness. \end{proof} \begin{theorem}[Typed Galois covariance of equality certificates] \label{thm:galois-certificate} Let $\sigma_t\in\operatorname{Gal}(K_k/\Q)$ be defined by $\sigma_t(\zetaK)=\zetaK^t$, where $\gcd(t,k)=1$. Transport a typed certificate $C$ by carrying every signed coefficient, role tag, and rotation tag at exponent $e$ to the residue $te\bmod k$; denote the resulting typed certificate by $D_tC$. Then \begin{equation} A_{D_tC}(\zetaK)=\sigma_t(A_C(\zetaK)),\qquad B_{D_tC}(\zetaK)=\sigma_t(B_C(\zetaK)). \label{eq:galois-coefficients} \end{equation} Whenever $A_C(\zetaK)\ne0$, its equality candidate transforms as \begin{equation} -\frac{B_{D_tC}(\zetaK)}{A_{D_tC}(\zetaK)} =\sigma_t\!\left(-\frac{B_C(\zetaK)}{A_C(\zetaK)}\right). \label{eq:galois-candidate} \end{equation} Thus the unfiltered set of typed realized equality candidates is closed under the Galois action. \end{theorem} \begin{proof} For an exponent residue $e$, the four role templates are \[ z^e,\qquad z^e+z^{-e},\qquad z^e-z^{-e},\qquad z^e+z^{-e}-2. \] Evaluation of the corresponding template at exponent $te$ is exactly the $\sigma_t$-image of its evaluation at $e$; the rational constant $-2$ is fixed. Multiplication by a unit permutes all residues for embedding and seed certificates and all nonzero residues for lower certificates, so the transport is again typed in the same role. Integer linearity gives \eqref{eq:galois-coefficients}, and division in $K_k$ gives \eqref{eq:galois-candidate}. \end{proof} This theorem is intentionally prior to the selector's order tests. Galois automorphisms need not preserve positivity: at $k=10$, $1+2\cos(2\pi/5)>1$ is carried by $\sigma_3$ to $1+2\cos(6\pi/5)<1$. Therefore no Galois invariance of the admissible $\rho>1$ slice, complete sign covector, or terminal MDS follows. General divisibility criteria and classifications of abstract root-of-unity relations~\cite{Mann1965,ConwayJones1976,LamLeung2000,KoleyArikatla2019,DeCarliLaporta2026} can now be applied as filters. They become word theorems only after their coefficient hypotheses are checked against this certificate image. \subsection{Rational twice-prime lower resonances} \label{sec:rational-classification} We now specialize to DNA, $k=2p$ with $p$ an odd prime, and rational $\rho>1$. The statement concerns only the lower form $\lambda_s$. In one prime-sine block, the two letters on the positive side and the two on the negative side contribute sums of two vertices of $V_2=\{\pm e_H,\pm e_L\}$. Under scalar evaluation \[ \omega_\rho(x_H,x_L)=\rho x_H+x_L, \] the possible nonzero absolute pair-sum values are \begin{equation} 2\rho,\qquad \rho+1,\qquad \rho-1,\qquad 2. \label{eq:pair-sum-values} \end{equation} \begin{lemma}[Unique rational block collision] \label{lem:rho-three} For rational $\rho>1$, two distinct vector sums in $V_2+V_2$ have the same scalar value under $\omega_\rho$ only when $\rho=3$. At $\rho=3$ the only new equality in absolute value is $\rho-1=2$. \end{lemma} \begin{proof} Apart from zero and sign, distinct scalar pair sums are the four expressions in~\eqref{eq:pair-sum-values}. Pairwise equality among them has no solution $\rho>1$ except $\rho-1=2$, which gives $\rho=3$. Equality to the negative of another positive value is impossible. Direct inspection of the signed pair sums proves the claim. \end{proof} At $\rho=3$, the exceptional ordered blocks, including exchange of the two sides, are generated by \begin{equation} (T,C),(C,T)\ \longleftrightarrow\ (G,G), \qquad (A,G),(G,A)\ \longleftrightarrow\ (C,C). \label{eq:exceptional-blocks} \end{equation} There are eight ordered side-aware patterns in total. \begin{theorem}[Rational twice-prime lower-resonance classification] \label{thm:rational-twice-prime} Let $p$ be an odd prime, $k=2p$, $h=(p-1)/2$, and let $\rho>1$ be rational. Then every realized lower zero is structural unless $\rho=3$. At $\rho=3$, a lower word is zero exactly when every prime-sine block is either structural or one of the eight exceptional blocks in \eqref{eq:exceptional-blocks}. It is accidental exactly when at least one block is exceptional. Consequently \begin{align} Z_{\mathrm{str}}(4,2p)&=4\cdot36^h, \label{eq:dna-zstruct}\\ Z_{\mathrm{tot}}(3;4,2p)&=4\cdot44^h, \label{eq:dna-ztotal}\\ Z_{\mathrm{acc}}(3;4,2p)&=4(44^h-36^h). \label{eq:dna-zacc} \end{align} The corresponding physical firing counts are \begin{align} F_{\rho\ne3}(4,2p)&=4^{2p-2}-36^h, \label{eq:dna-fgeneric}\\ F_{\rho=3}(4,2p)&=4^{2p-2}-44^h. \label{eq:dna-fthree} \end{align} \end{theorem} \begin{proof} By \Cref{lem:prime-sines}, the full lower form vanishes exactly when the scalar coefficient of every prime-sine block vanishes. If the two vector sums in a block are equal, the block is structural. If they differ but have equal scalar values, \Cref{lem:rho-three} forces $\rho=3$, and direct use of the signed one-hot certificate gives exactly \eqref{eq:exceptional-blocks}. This proves the classification without forgetting the realizing letters. For DNA, \Cref{lem:cross-energy} gives $36$ structural ordered assignments per block. At $\rho=3$ there are eight additional assignments, hence $44$. The $h$ blocks are independent and the central letter is free, giving \eqref{eq:dna-zstruct} and \eqref{eq:dna-ztotal}; subtraction gives \eqref{eq:dna-zacc}. Finally apply the quartering argument of \Cref{thm:twice-prime-structural} to the total zero count, yielding \eqref{eq:dna-fgeneric} and \eqref{eq:dna-fthree}. \end{proof} For $p=3$, the theorem gives $144$ structural zeros, $176$ total zeros at $\rho=3$, and $32$ accidental zeros. The firing counts are respectively $220$ and $212$. These numbers are checks of the theorem, not its proof. The tempting ratio $\rho=2$ is not exceptional: the four positive quantities in~\eqref{eq:pair-sum-values} are then $4,3,1,2$, all distinct. \begin{corollary}[Scoped twice-prime DNA trichotomy] \label{cor:twice-prime-dna-trichotomy} Fix $k=2p$ with $p$ an odd prime and consider only the DNA \role{lower-lambda} equality family. Then: \begin{enumerate} \item if $\rho\notin K_{2p}^+$, every lower zero is structural; \item if $\rho>1$ is rational, an accidental lower zero exists if and only if $\rho=3$; and \item among ratios not decided by these two clauses, only $(K_{2p}^+\setminus\Q)\cap(1,\infty)$ remains. \end{enumerate} The third clause is an exact residual domain, not an existence assertion. \end{corollary} \begin{proof} The first clause is the necessary field gate of \Cref{cor:dna-field-gate}. The second is \Cref{thm:rational-twice-prime}. Partitioning the admissible real ratios $\rho>1$ by field membership and rationality leaves precisely the set in the third clause; field membership alone remains only necessary. \end{proof} \subsection{Exact direct-membership complexity} \label{sec:complexity} The direct formula~\eqref{eq:terminal-selector} avoids materializing the MDS, but an exact complexity statement requires a literal representation of every real algebraic input. We use the following model. Let a real algebraic number $\alpha$ be supplied by a square-free polynomial $f\in\Z[x]$, normalized to have positive leading coefficient, and a rational isolating interval selecting $\alpha$. Put $e=\deg f$. Each positive magnitude is represented by a dense coefficient list as \[ \beta_i=\frac{b_i(\alpha)}{d_i},\qquad b_i\in\Z[x],\quad \deg b_i0}. \] The complement table, a total alphabet order, even $k\ge4$, and the queried word are explicit. Let $L$ denote the total bit length of this input. Degrees and every coefficient are encoded explicitly; in particular $e=O(L)$. \begin{theorem}[Dense variable-degree polynomial-bit membership] \label{thm:membership-complexity} Preprocessing time, per-word query time, and working memory for membership in the direct MDS of \Cref{thm:imported-selector} are polynomial in $k$, $q$, and $L$, even though $e$ is part of the input. The algorithm is exact and stores no table of size $q^{k-1}$ or $q^k$. \end{theorem} \begin{proof} Put $c=2\cos(2\pi/k)$. Its defining polynomial and an isolating interval can be computed from $\Phi_k$ in polynomial bit complexity in $k$: the degree is at most $\varphi(k)/2\le k/2$, and the required coefficient bit sizes are polynomial in $k$. Write $\mu_k\in\Z[y]$ for the square-free minimal polynomial selecting $c$ and $\delta=\deg\mu_k$; thus $\delta=\varphi(k)/2$ and $\mu_k$ is monic. The recurrences \[ 2\cos(j\theta)=C_j(c),\qquad \frac{\sin(j\theta)}{\sin\theta}=U_{j-1}(c/2), \qquad \theta=2\pi/k, \] where $C_0(x)=2$, $C_1(x)=x$, and $C_{j+1}(x)=xC_j(x)-C_{j-1}(x)$; equivalently, $C_j(x)=2T_j(x/2)$ for the standard Chebyshev polynomial $T_j$. Express every real, normalized imaginary, lower-lambda, and midpoint branch as \begin{equation} \operatorname{sgn} G(\alpha,c) \label{eq:bivariate-sign} \end{equation} after multiplication by known positive denominators. Here $G\in\Z[x,y]$ has degree less than $e$ in $x$, degree $O(k)$ in $y$, and coefficient bit length polynomial in $k+L$. These bounds follow by the Chebyshev recurrences and by summing at most $k$ represented weights. Before invoking the sign routine, reduce $G$ in the two algebraic coordinates. Pseudo-division in $x$ by $f$ replaces $G$ by an integral remainder of $x$-degree less than $e=\deg f$, after multiplication by a known positive power of the leading coefficient of $f$. Division in $y$ by the monic polynomial $\mu_k$ then gives an integral representative $\widetilde G$ of $y$-degree less than $\delta$. These operations preserve the value sign at $(\alpha,c)$, including zero. The two reductions use a number of dense integer-polynomial operations polynomial in $e+k$. Standard pseudo-division height bounds, resultant construction of $\mu_k$, square-free factorization, and real-root isolation keep all degrees and coefficient bit lengths polynomial in $e+k+L$, hence in $k+L$. Put $d_2=\max(e,\delta)$ and $d_1=\max(\deg_x\widetilde G,\deg_y\widetilde G)$; then $d_11$, an accidental DNA embedding zero exists if and only if \[ \begin{cases} a+b\le2^{t+1},&a,b\text{ both odd},\\ a+b\le2^t,&a,b\text{ of opposite parity}. \end{cases} \tag{F} \] Equivalently, \[ \boxed{\E_t=\left\{\frac{v+u}{v-u}:00$. Then $c\cdot\beta=\sum_j\delta_j\sum_{i=j}^m c_i$. This has a zero in the open positive cone precisely under the stated mixed-sign condition. A distinct intersecting hyperplane has relatively open points not covered by the finitely many other hyperplanes. The rational structure also supplies rational points there. \end{proof} The equivalence to the cyclotomic embedding floor here assumes rational magnitudes. The additive arrangement itself is defined for real magnitudes, but it must not be substituted for all algebraic cyclotomic equalities. \paragraph{A complete three-magnitude instance.} For Freiman order $2$ and $01$, $v=z/x>u$. Their interior intersections are $(2,3),(3,5),(3,7),(3,9)$, of multiplicities $2,3,2,2$. Adding lines inside the convex cone gives \[ 1+6+\bigl[(2-1)+(3-1)+(2-1)+(2-1)\bigr]=12 \] open chambers. There are $6+(2+3+2+2)=15$ one-dimensional faces and four vertices. This is an exact additive-floor arrangement (and, by the preceding theorem, the rational lower-floor arrangement at $k=2p,q=6$), not the full terminal arrangement. \subsection{Exact embedding counts with two odd primes}\label{ext:sec:counts} Let $n(d)$ be the number of ordered pairs in $V_m^2$ with difference $d$. For structural counting retain vector $d$. For rational weights aggregate multiplicities after scalar projection. \begin{theorem}[Rectangular counting formula]\label{ext:thm:count} At $k=2p\ell$, with two distinct odd primes, the corresponding number of zero words is \[ Z_{p,\ell}(n)= \sum_{(b_1,\ldots,b_\ell)\in D_m^\ell} \left(\prod_{j=1}^{\ell}n(b_j)\right) \left(\sum_{a\in D_m-D_m}\prod_{j=1}^{\ell}n(a+b_j)\right)^{p-1}, \tag{C} \] where $n$ is extended by zero outside its support. For scalar counting the alphabet and difference set in (C) are their projected versions. At $k=2^bp^a\ell^c$ the count is the $(k/(2p\ell))$th power of (C). \end{theorem} \begin{proof} The two-dimensional anchored-difference kernel consists exactly of arrays $c_{ij}=a_i+b_j$. Fix the gauge $a_1=0$, so $b$ is the actual first row and is unique. Each other row independently chooses its unique translating value $a_i$; its number of letter-pair realizations is the product of entry multiplicities. Any feasible shift lies in $D_m-D_m$. Multiply row weights and sum. The rational prime-power block decomposition is independent in each of the $k/(2p\ell)$ position classes, giving the power. \end{proof} For DNA at $k=30$ this gives: \begin{center}\begin{tabular}{lrr}\toprule Ratio & Total embedding-zero words & Accidental words\\\midrule Structural floor & 21,566,097,316 & 0\\ $5/3$ & 21,573,498,016 & 7,400,700\\ $2$ & 27,518,030,956 & 5,951,933,640\\ $3$ & 288,112,407,796 & 266,546,310,480\\ $5$ & 22,778,390,836 & 1,212,293,520\\ $7$ & 21,566,157,256 & 59,940\\\bottomrule \end{tabular}\end{center} These values were independently evaluated by first-row enumeration and first-row histograms; no enumeration of $4^{30}$ words was used. \paragraph{One odd prime.} For DNA at $k=2p$, structural coefficients must be constant on the prime block. The multiplicities of $D$ are $n(0)=4$, four axis differences with multiplicity $1$, and four diagonals with multiplicity $2$. Thus \[ Z^{\rm str}_{2p}=4^p+4+4\,2^p,\qquad Z_{2p}(3)=4^p+2+2\,2^p+2\,3^p. \] At $2^bp^a$ raise these to the $2^{b-1}p^{a-1}$th power. At other rational ratios the structural count applies. \paragraph{PCRs at every even order.} Every word of proper period $d\mid k$ has componentwise zero embedding, because the geometric sum across repeats vanishes. Burnside therefore gives, for either structural or fixed-ratio total zero-word count $Z_k$, \[ \#\{\text{zero PCRs}\}=\frac1k\left(Z_k+\sum_{d\mid k,\ d0$. The largest ratio is \[ \max\mathcal R_p=2p+4-(-1)^{(p-1)/2}. \] There are exactly four structural midpoint words, and all four have $\lambda=0$. \end{theorem} \begin{proof} First assume $\rho\notin\E_2$. Equality of the scalar values of two $C_j$ forces equality of their vectors: a nonzero difference is an even-norm vector of norm at most $8$, whose possible rational annihilators are exactly $\E_2$. Thus all $C_j=C$. For $C=0$, the vector $\sum D_j/2+s$ is any sum of $2h+1=p$ signed units, giving $\mathcal R_0$. For $C=(2,0)$, divide the defect in (M) by $2$; it becomes $(1+2a,2b)$ where $(a,b)$ is a sum of $h+1$ units, giving $\mathcal R_{20}$. For $C=(1,1)$, write \[ \sum D_j=u(1,1)+2v(1,-1),\quad u=-h,-h+2,\ldots,h,\quad -h\le v\le h. \] The choices of $u,v$ are independent by the table. Taking the central letter to be $\epsilon e_1$, and using interchange symmetry for $\epsilon e_2$, gives $\mathcal R_{11}$. For $C=(3,1)$, the sum of the $D$ values is $t(1,-1)$, with $t\equiv h\pmod2$ and $|t|\le h$. Two central signs give ratio $1$ and are excluded; the other two give $\mathcal R_{31}$. For $C=(2,2)$, the sum is $2t(1,-1)$ for any integer $|t|\le h$, giving $\mathcal R_{22}$. The type $(4,0)$ gives either a structural zero or no ratio greater than one. Every listed choice is reversible into its constituent ordered letter pairs, proving sufficiency on actual words. The excluded exceptional set $\E_2$ adds nothing. Ratio $2$ is in $\mathcal R_{20}$: use $(a,b)=(0,1)$ for even $h$, and $(-1,1)$ for odd $h$. Ratios $3,5$ are in $\mathcal R_{22}$ using $t=0,1$. Ratios $5/3,7$ are in $\mathcal R_{31}$ using $t=0$ for even $h$ and $t=-1$ for odd $h$. Thus every possible exceptional ratio is already in the proposed union. For structural vanishing, the vector analogue of (M) holds with common $C$. Since each coordinate of $C-D_j$ is even, $C+2\sum D_j+4s\equiv pC\pmod4$. Hence each coordinate of $C$ is divisible by four. The type $C=0$ cannot work: it would make an odd number $p$ of signed units sum to zero. Only $C=\pm4e_i$ remain, with all $D_j=0$ and $s=\mp e_i$. They yield exactly four words and each symmetric pair has identical letters, so its lambda contribution is zero. To make a nonstructural midpoint equality active, it suffices that some symmetric pair has distinct letters. Swapping those two positions leaves $C,D$, and hence midpoint zero, unchanged; it changes lambda by a nonzero real multiple of their weight difference. At least one of the two words has nonzero lambda; reversal changes its sign while preserving midpoint, yielding positive lambda. Types $20,11,31$ have such a pair in every realization. The set $\mathcal R_0$ is contained in $\mathcal R_{20}$: orient the odd coordinate as $1+2a$, and use its two possible signs to obtain the required parity with $|a|+|b|\le h+1$. For type $22$, if $|t|0$. Therefore the twice-prime structural-tie inactivity statement must not be extended to all even orders. The general two-family theorem retains these ties. \end{remark} \subsection{Every real cyclotomic ratio is eventually realizable}\label{ext:sec:algebraic} \begin{theorem}[Eventual positional realization]\label{ext:thm:algebraic} A real algebraic number $\rho>1$ is realized as an accidental DNA embedding ratio at some even order if and only if it belongs to a cyclotomic field. For every representation $\rho=U(\zeta_n)/d$, with even $n\ge4$, $d\in\Z_{>0}$ and $U\in\Z[X]$ reduced to degree less than $n$, there is an explicit one-hot construction at \[ k=n\ell_1\cdots\ell_t, \] where the $\ell_i$ are fresh distinct odd primes and $2^t\ge\max_e\norm{(2d\,\mathbf1_{e=0},-2U_e)}$. \end{theorem} \begin{proof} Necessity is the preceding field gate. For sufficiency take $t\ge1$ satisfying the bound. For each $e=0,\ldots,n-1$, the vector $c_e=(2d\mathbf1_{e=0},-2U_e)$ has even norm at most $2^t$, so express it as a sum of exactly $2^t$ signed units, padding with opposite pairs. Place these units, with alternating signs, at the corners of a binary $t$-cube. Extend to the full $\ell_1\times\cdots\times\ell_t$ grid by collapsing all nonanchor values to one corner index, exactly as in Lemma~\ref{ext:lem:realization}, but using individual letters rather than letter pairs. The prime-root sum in each new coordinate reduces this array to a nonzero factor $\Gamma$ times $c_e$. CRT identifies all $(e,j_1,\ldots,j_t)$ with distinct positions in the length-$k$ word; every position receives one letter. The resulting component embeddings are \[ H=2d\Gamma\ne0,\qquad L=-2\Gamma U(\zeta_n), \] so $\rho H+L=0$ and the zero is accidental. Explicitly one may take the exponent $\frac{k}{n}e+\sum_i\frac{k}{\ell_i}j_i\pmod k$; all coefficients of this CRT map are units in their own factors. No abstract coefficient is left without a positional realization. \end{proof} The cyclotomic-field gate is therefore sufficient \emph{after allowing the order to grow}, although it remains insufficient at a fixed order. Examples verified by exact polynomial division are $\rho=\sqrt2$, base $n=8$, fresh prime $3$, giving $k=24$; and $\rho=(7+3\sqrt5)/2$, base $n=10$, fresh primes $3,7,11$, giving $k=2310$. These are explicit upper bounds, not minimum realizing orders. \begin{corollary}[Eventual raw realization for the other roles] Over all even orders, the same necessary and sufficient cyclotomic-field condition holds for the raw real, imaginary, lower-lambda and midpoint equality families. \end{corollary} \begin{proof} For the real and imaginary roles, rotate an embedding witness until the relevant component of its nonzero $H$ is nonzero. Such rotations exist because $k\ge4$. Delete the exponent-zero letter for a lower-lambda witness; its imaginary contribution was zero. For midpoint, interleave the embedding witness with its letterwise complement at order $2k$. This gives component weight sum zero and new component embedding $(1-\zeta_{2k})H\ne0$, while its scalar embedding is zero. Rotate to make its real high component nonzero and delete the exponent-zero letter; the midpoint form does not see that position. Necessity follows by solving the corresponding nonstructural linear relation in the real cyclotomic coefficient field, after the usual positive sine normalization for lambda. \end{proof} The midpoint witnesses in this corollary have scalar lambda zero. It does \emph{not} establish eventual active midpoint realization for every algebraic ratio. Fixed-order realization, efficient/minimal conductors, and active branch classification remain separate problems. \paragraph{Uniform radical designs.} The preceding uniform magnitude family is linearly independent over every cyclotomic coefficient field. Since all catalogue forms, not just embedding and normalized lambda, have coefficients in such fields, the same componentwise argument makes every zero in the entire catalogue structural. This does not remove structurally active midpoint ties such as the $k=8$ example. \subsection{Residual geometry as a three-block problem}\label{ext:sec:paths} The imported construction proves strict increase of the midpoint potential along arcs between positive-lambda states \cite{Babanskyy2026Parity}. The following block interface and max-plus formula are deductions, not a new claim to that monotonicity lemma. Set \[ \begin{gathered} N=\{u:\lambda_s(u)\le0\},\\ P_0=\{u:\lambda_s(u)>0,a_s(u)=0\},\qquad P_1=\{u:\lambda_s(u)>0,a_s(u)=1\}. \end{gathered} \] \begin{theorem}[Three-block residual interface]\label{ext:thm:paths} There is no original lower-graph arc from $P_0$ to $P_1$. Every residual arc respects the order \[ P_0\longrightarrow N\longrightarrow P_1. \] The positive blocks contain all their original internal arcs. The core is the original graph induced on $N$ with only the zero-PCR seed arcs deleted. RC swaps the two positive blocks, reverses residual arcs and preserves the core. Let $L_P$ be either positive-block longest path, $L_N$ the core longest path, and let $d_P(u)$ be the longest $P_0$-internal path ending at $u$. Define \[ \eta(n)=\max\bigl(\{0\}\cup\{d_P(u)+1:u\in P_0,\ u\to n\}\bigr). \] Then the exact residual longest path is \[ \boxed{L(D_s)=\max\left\{L_P,\ \max_{n_0\to\cdots\to n_j\text{ in }N} [\eta(n_0)+j+\eta(\RC(n_j))]\right\}.} \tag{P} \] In particular $L_N\le L(D_s)\le2L_P+L_N+2$. \end{theorem} \begin{proof} For any lower arc $u\to v$, \[ r_s(v)-r_s(u)=\tan(\pi/k)(\lambda_s(u)+\lambda_s(v)). \] This follows by writing $P=x+iy$ in $(\zeta_k^{-1}-1)P$. If both lambdas are positive, the increment is strictly positive. A $P_0$ state has $r\ge0$, and a $P_1$ state has $r\le0$, so $P_0\to P_1$ is impossible. The two-family formula now checks all block types: $N\to P_0$, $P_1\to N$ and $P_1\to P_0$ have terminal indicator one; $P_0\to N$ and $N\to P_1$ have indicator zero. On each positive block it is zero, and on $N$ it is the zero-PCR seed indicator. The imported RC invariance gives the stated anti-isomorphism. A residual path either lies inside a positive block or decomposes into a $P_0$ prefix, one core path, and a $P_1$ suffix; either end may be empty. For a fixed core path the longest possible prefix is $\eta(n_0)$. The RC image of a longest incoming prefix at $\RC(n_j)$ is a longest outgoing suffix at $n_j$, with length $\eta(\RC(n_j))$. These paths lie in disjoint blocks, so they concatenate without collision. Maximizing proves (P). The inequalities follow from $\eta\le L_P+1$. \end{proof} Formula (P) can be evaluated by a weighted DAG dynamic program on the core after computing the two boundary functions; it does not claim subexponential enumeration of all lower states. \begin{center}\begin{tabular}{llrrr}\toprule $k$ & Weight or interval & $L_P$ & $L_N$ & $L(D_s)$\\\midrule 4 & any atlas cell &2&11&17\\ 6 & first atlas interval &9&43&63\\ 6 & $\rho=3$ &8&47&65\\ 6 & $3<\rho<4$ &9&39&59\\ 6 & $4<\rho<5$ &8&39&57\\ 8 & $\rho=2$ &33&79&147\\ 8 & $\rho=(7+3\sqrt5)/2$ &28&85&143\\ 8 & $\rho=\sqrt{13}$ &27&85&141\\ 8 & $\rho=\sqrt[3]{2}$ &33&93&161\\\bottomrule \end{tabular}\end{center} All 199 certified strata at $k=4,6,8$ satisfy equality in the upper bound, but we assert only the exact formula (P) and the inequality at arbitrary order. An all-order equality claim would need an additional argument aligning extremal prefixes, core paths, and suffixes. \subsection{Minimal root locations, positivity, and finite optimization}\label{ext:sec:remaining} \begin{proposition}[Minimal one-parameter cut set] Let a complete finite affine decision catalogue on an open interval be given, and evaluate the deterministic output on every point and interval cell. A root is indispensable to a point/interval partition exactly when the outputs on its left interval, at the point, and on its right interval are not all equal. Keeping exactly these roots gives the unique minimum set of root locations for such a partition. With $E$ indispensable roots it has $2E+1$ cells; the number of distinct outputs can be smaller. \end{proposition} \begin{proof} If the three outputs agree, removing that root merges cells with identical output. Conversely, without an indispensable root the output would have to be constant in a neighborhood containing all three local cells, a contradiction. Finite repeated deletion cannot change this local necessity. The count is $E$ points and $E+1$ open intervals. A point-only anomaly must therefore be retained even when its adjacent outputs agree. \end{proof} Applied to the complete small atlases: $k=4$ has one indispensable root, three cells and two distinct terminals; $k=6$ has eleven indispensable roots, twenty-three cells and seventeen terminals. This is not a general solution for minimum forms in higher-dimensional weight space. The complete $k=8$ atlas is given in \Cref{thm:complete-atlas}. \begin{proposition}[Two barriers to terminal Galois transport] A field automorphism of a real number field preserving positivity of every element is the identity. A fixed coordinate permutation on words over an alphabet with at least two letters preserves all directed de Bruijn arcs only when it is the identity; it reverses all arcs only when it is position reversal. \end{proposition} \begin{proof} If $x\ne\sigma(x)$, a rational strictly between them supplies an element whose sign changes under $\sigma$. For the graph statement, write the image of a word as $(w_{\pi(0)},\ldots,w_{\pi(k-1)})$. In an arc $w\to(w_1,\ldots,w_{k-1},a)$, the free appended letter cannot enter an overlap constraint. Those constraints then force $\pi(j+1)=\pi(j)+1$ for all $j$, hence the identity. Compose an arc-reversing permutation with position reversal to obtain the first case. \end{proof} These barriers do not exclude a finite set of certificates retaining positivity, accidental terminal coincidences, or alphabet symmetries. They explain why equality-level Galois covariance alone cannot be promoted to a terminal orbit classification. \paragraph{An explicit, bounded optimization result.} Choose the projective angular coordinate $\theta=\arctan\rho\in(\pi/4,\pi/2)$, and define robustness as distance to all critical ratios and the two domain boundaries in this coordinate. First minimize residual longest path in the complete atlas, then maximize robustness inside an optimal open cell. At $k=4$, the widest optimal cell is $(1,3)$; its angular midpoint is \[ \rho=(1+\sqrt5)/2. \] At $k=6$, the minimum is $L(D)=57$ for $\rho\ge4$. Among the optimal open cells, $(11,\infty)$ is angularly widest, giving \[ \rho=11+\sqrt{122}. \] For example its width is $\arctan(1/11)$; the other finite optimal-cell widths are $\arctan(1/21)$, $\arctan(1/18)$, $\arctan(1/32)$, and $\arctan(1/50)$, all smaller. This is an exact answer for a declared metric and two complete small atlases, not a universal or biologically optimal weight. The first ratio is the ordinary golden ratio, not the earlier landmark $\varphi^4$.