\section{Reciprocal evaluation and the lattice of false positives}\label{sec:lattices} Put $d=\varphi(t)$, $Q=\Phi_t(B)$, and choose a primitive $t$-th root $\zeta$. Write $\A=\Z[\zeta]\simeq\Z[X]/(\Phi_t(X))$. We use the classical facts that cyclotomic polynomials are monic and irreducible, have constant term 1 for $t>1$, and are reciprocal for $t>2$; see the opening conventions of~\cite[\S1]{Reddy}. The norm of a nonzero element is the determinant of its multiplication map in the integral basis $1,\zeta,\ldots,\zeta^{d-1}$. \begin{lemma}[Reciprocal evaluation and end digits]\label{lem:endpoints} The base $B$ is invertible modulo $Q$, and \begin{equation}\label{eq:double} Q\mid N,N^{\rm rev}\quad\Longleftrightarrow\quad P(B)\equiv P(B^{-1})\equiv0\pmod Q. \end{equation} A false positive can be extended to any greater length by adding zeros at the most significant end. Every minimum false positive has nonzero end digits. To prove $K_t(B)=K$, it is therefore enough to exclude the whole box at $K-1$ and construct a false positive at $K$. \end{lemma} \begin{proof} Since $\Phi_t(0)=1$, we have $\gcd(B,Q)=1$. Equation~\eqref{eq:digits} gives \eqref{eq:double} after cancelling the unit $B^{k-1}$. Reciprocity ensures that both evaluations vanish on the divisor and descend to $\A$. Adding a most significant zero leaves $N$ unchanged and multiplies $N^{\rm rev}$ by $B$. If the least significant digit is zero, write $P=XP_1$: removing $X$ multiplies the algebraic value by $\zeta^{-1}$ and cancels only units in the modular evaluations. A most significant zero can be removed directly. Either removal would preserve a shorter false positive. \end{proof} \begin{proposition}[Image of reciprocal evaluation]\label{prop:image} For $\delta=\gcd(Q,B^2-1)$, the additive map \[ E:\A\longrightarrow(\Z/Q\Z)^2,\qquad f(\zeta)\longmapsto(f(B),f(B^{-1})) \] has image exactly $\{(u,v):u\equiv v\pmod\delta\}$. In particular, \begin{equation}\label{eq:image-index} [\A:\ker E]=Q^2/\delta. \end{equation} \end{proposition} \begin{proof} Every difference $f(B)-f(B^{-1})$ is a multiple of $B-B^{-1}$ modulo $Q$. The ideal generated by this element is generated by $\delta$, since multiplication by $B$ turns it into $B^2-1$. Conversely, if $\delta\mid u-v$, choose $b$ with $b(B-B^{-1})\equiv u-v\pmod Q$ and put $a=u-bB$. Then $a+bX$ realizes $(u,v)$. There are $Q(Q/\delta)$ such pairs. \end{proof} \begin{proposition}[Intersection and product]\label{prop:ideals} Let \[ I=(B-\zeta)\A,\qquad J=(B-\zeta^{-1})\A,\qquad\Lambda=I\cap J, \] and \begin{equation}\label{eq:H} H=(B-\zeta)(B-\zeta^{-1})=B^2+1-B(\zeta+\zeta^{-1}). \end{equation} Then \begin{equation}\label{eq:indices} \Lambda=\ker E,\qquad H\A\subseteq\Lambda,\qquad [\A:H\A]=Q^2,\qquad[\Lambda:H\A]=\delta. \end{equation} For every complex embedding $\sigma$, $\sigma(H)=B^2+1-2B\cos\theta\ge(B-1)^2>0$. \end{proposition} \begin{proof} The quotient by $I$ is $\Z/Q\Z$, with $\zeta\mapsto B$. In the quotient by $J$ we have $B\zeta=1$; reciprocity again gives $\Z/Q\Z$, now with $\zeta\mapsto B^{-1}$. Hence $\Lambda=\ker E$. The product $IJ=H\A$ is contained in the intersection. Both $B-\zeta$ and $B-\zeta^{-1}$ have norm $Q$, so multiplication by $H$ has determinant $Q^2$. The index formula and \eqref{eq:image-index} give the last index. The final inequality follows from $\cos\theta\le1$. \end{proof} \begin{remark}[Retaining every coset] Every $\alpha\in\Lambda$ has the form $H\beta$ with $\beta\in\A$ if and only if $\delta=1$. For $\delta>1$, that parametrization covers only the sublattice $H\A$. Section~\ref{subsec:fifths} explicitly constructs the missing cosets at order ten. \end{remark} \begin{proposition}[Classification of $\delta_t(B)$]\label{prop:delta} For $t\ge3$, \begin{equation}\label{eq:delta} \delta_t(B)=\begin{cases} p,&t=p^a,\ p\text{ odd prime},\ a\ge1,\ B\equiv1\pmod p,\\ p,&t=2p^a,\ p\text{ odd prime},\ a\ge1,\ B\equiv-1\pmod p,\\ 2,&t=2^a,\ a\ge2,\ B\text{ odd},\\ 1,&\text{otherwise}. \end{cases} \end{equation} \end{proposition} \begin{proof} Evaluating $(X^n-1)/(X-1)=\prod_{e\mid n,e>1}\Phi_e(X)$ at 1 gives $\prod_{e\mid n,e>1}\Phi_e(1)=n$. Induction yields $\Phi_n(1)=p$ for $n=p^a$, and 1 otherwise: when $n$ has at least two prime factors, the proper prime-power factors already supply the whole product $n$. The identities $\Phi_{2m}(X)=\Phi_m(-X)$ for odd $m>1$, and evenness of the polynomial when $4\mid n$, give, for $n\ge3$, \[ \Phi_n(-1)=\begin{cases}p,&n=2p^a,\ p\text{ odd},\\2,&n=2^a,\ a\ge2,\\1,&\text{otherwise}.\end{cases} \] For odd $n$ one can also use $\Phi_n(-1)=\Phi_{2n}(1)=1$. An odd prime dividing $B^2-1$ has $B\equiv\pm1$ and must divide one of these two values. Only the first two cases of \eqref{eq:delta} can occur. To exclude higher powers, in the first case put $C=B^{p^{a-1}}=1+ph$. Modulo $p^2$, \[ \Phi_{p^a}(B)=\sum_{j=0}^{p-1}C^j \equiv p+ph\sum_{j=0}^{p-1}j\equiv p\pmod{p^2}. \] Replace $B$ by $-B$ for the second case. For the prime 2, an even base contributes no factor to $B^2-1$. For odd $B$, $\Phi_t(B)\equiv\Phi_t(1)\pmod2$, so $t$ must be a power of 2. If $t=2^a$, $a\ge2$, then $\Phi_t(B)=B^{2^{a-1}}+1\equiv2\pmod8$. The factor is exactly 2. \end{proof}