\section{The coherent factor and its transpose pencil} \label{sec:coherent-transpose} From here to the end of Section~\ref{sec:conjugate-gluing}, $R$ is an Osserman algebraic curvature tensor on $\R^{16}$ with a reduced eigenvalue $\mu$ of multiplicity seven or eight that is neither the smallest nor the largest reduced eigenvalue (the extreme case is Theorem~\ref{thm:extreme}). Shift $\mu$ to zero; the shifted Jacobi operators then have rank $\nu\in\{7,8\}$; put $k=16-\nu$. Let $\Lambda$ be the diagonal matrix of the nonzero eigenvalues, with multiplicities; it is indefinite because $\mu$ is neither the smallest nor the largest reduced eigenvalue. Put $H=\Lambda^{-1}$ and $\mathcal B(X,Y)=\tfrac12(\Jac_{X+Y}-\Jac_X-\Jac_Y)$. For a real unit anchor $X$ choose orthonormal frames $M_X:\R^\nu\to E_X=\operatorname{Im}\Jac_X$ and $K_X:\R^k\to\ker\Jac_X$, with $\Jac_X=M_X\Lambda M_X^T$. A \emph{regular anchor} is a point of the nonempty Euclidean-open set $U_0$ constructed in Section~\ref{sec:assembly}: the generic-pair construction below applies there, and $U_0$ avoids the proper algebraic exceptional loci named in the following sections. Section~\ref{sec:assembly} partitions $U_0$ into finitely many semialgebraic sets $\mathcal U_j$ on which all pole invariants are constant (Lemma~\ref{lem:asm-partition}); we call them \emph{branches}. It also lists the cases, (L1)--(L8), and names the section that settles each; a first reading may start there. \subsection{Construction and normalization} Take $X$ in the nonempty open set of anchors where the generic-pair construction of Lemma~\ref{lem:rationalpartner} applies (only the two-plane result of \cite[Section~4]{Nik2005} is used, not its three-plane construction, whose dimension hypothesis fails here). As in that proof, $M_Y=M_XF(Y)+K_XP(Y)$ with \begin{equation}\label{eq:cp-lower} P(Y)=K_X^T\,2\mathcal B(X,Y)M_X\Lambda^{-1} \end{equation} linear, $F$ rational and homogeneous of degree one, and the block identities \begin{align} F\Lambda+\Lambda F^T&=M_X^T\,2\mathcal B(X,Y)M_X,\label{eq:cp-mixed}\\ P\Lambda F^T&=K_X^T\Jac_Y M_X,\label{eq:cp-cross}\\ F\Lambda F^T&=M_X^T\Jac_Y M_X,\qquad P\Lambda P^T=K_X^T\Jac_Y K_X\label{eq:cp-diagonal} \end{align} hold as rational identities; $P$, which we call the \emph{lower pencil}, is injective for generic $Y$. Thus \begin{equation}\label{eq:cp-global-factor} M_{sX+Y}=sM_X+M_XF(Y)+K_XP(Y),\qquad \Jac_{sX+Y}=M_{sX+Y}\Lambda M_{sX+Y}^T. \end{equation} \begin{lemma}\label{lem:cp-pole-free} If $F$ has no pole along any complex hypersurface in $X^\perp$, then \eqref{eq:cp-global-factor} is a global linear factor, and the tensor is Clifford. \end{lemma} \begin{proof} In the polynomial ring on $X^\perp$, a rational function with no divisorial pole is polynomial; by homogeneity $F$ is linear, and \eqref{eq:cp-global-factor} is a polynomial identity in all sixteen variables. Its factor is injective at every nonzero real vector, as the Jacobi operator has rank $\nu$, so Proposition~\ref{prop:gramconditional}(1) applies. \end{proof} \subsection{Kernel directions give Jacobi factors} Extend $P$ linearly to $V=\R^{16}$ by $P(X)=0$. For $g\in\R^k$ define $N_g:\R^\nu\to V$ by \begin{equation}\label{eq:cp-N} N_g^T Y=P(Y)^Tg\qquad(Y\in V), \end{equation} linear in $g$, and for $a\in\R^\nu$ define the \emph{transpose pencil} $B(a):\R^k\to V$ by $B(a)g=N_ga$. \begin{lemma}\label{lem:cp-L20} For every real $g$, \begin{equation}\label{eq:cp-kernel-factor} \Jac_{K_Xg}=N_g\Lambda N_g^T. \end{equation} If $g\ne0$, then $N_g$ is injective. If $a\ne0$ is real, then $B(a)$ is injective; equivalently $Y\mapsto P(Y)a$ is onto $\R^k$. There are no nonzero fixed real coefficient space $U$ and proper fixed real subspace $T\subset\R^k$ with $P(Y)U\subset T$ for all $Y$. \end{lemma} \begin{proof} For $z=K_Xg$ and $v=sX+Y$ with $Y\perp X$, the relations $M_X^Tz=0$ and $P(X)=0$ give $M_v^Tz=P(Y)^Tg=N_g^Tv$ for every real $s$. So the coherent factor and the symmetry $\langle\Jac_vz,z\rangle=\langle\Jac_zv,v\rangle$ give \[ \langle\Jac_zv,v\rangle=(N_g^Tv)^T\Lambda(N_g^Tv)\qquad(v\in\R^{16}); \] polarizing in $v$ proves \eqref{eq:cp-kernel-factor}. Its left side has rank $\nu$ for real $g\ne0$, so $N_g$ is injective, and $N_ga=0$ with $a\ne0$ forces $g=0$; transposition gives surjectivity. Finally, for nonzero $g\perp T$ the containment would make $N_g$ vanish on $U$. \end{proof} Let $g_X$ be the unit vector with $K_Xg_X=X$. Then \begin{equation}\label{eq:cp-radial-skew} B(a)g_X=-M_Xa,\qquad K_X^TB(a)\text{ is skew-symmetric}, \qquad\operatorname{Im}B(a)\subset X^\perp. \end{equation} Indeed, injectivity of the coherent factor and $\Jac_zz=0$ give $M_z^Tz=0$; the coefficient of $s$ for $z=sX+Y$ gives $P(Y)^Tg_X=-M_X^TY$, the first identity. The last follows from $P(X)=0$. For the middle one, \eqref{eq:cp-kernel-factor} and $\Jac_{K_Xg}K_Xg=0$ give $N_g^TK_Xg=0$, so $\langle g,K_X^TB(a)g\rangle=0$ for all real $g,a$; polarize in $g$. We call the first identity in \eqref{eq:cp-radial-skew}, in either form, the \emph{radial identity}; $B(a)g_X=-M_Xa$ is the \emph{radial column} and $g_X^TP(Y)=-Y^TM_X$ the \emph{radial row}. For $\nu=8$ (so $k=\nu$) we say that the transpose pencil is \emph{aligned} at $X$ if $\operatorname{Im}B(a)=E_X$ for every real $a\ne0$. \begin{lemma}\label{lem:cp-complex-ranks} For every nonzero complex coefficient $a$, the kernel of $B(a)$ is totally isotropic in $\mathbb C^k$ for the standard bilinear form, and $s(a)=\rank B(a)$ satisfies $s(a)\ge\lceil k/2\rceil$ and $\rank(B(a)B(a)^T)=2s(a)-k$. \end{lemma} \begin{proof} If $0\ne g\in\ker B(a)$ with $g^Tg\ne0$, the Osserman characteristic and minimal-polynomial identities, continued to complex $g$, make $\Jac_{K_Xg}$ diagonalizable of rank $\nu$, so by \eqref{eq:cp-kernel-factor}, continued polynomially in $g$, $N_g$ has rank $\nu$, contrary to $N_ga=0$. So $L=\ker B(a)$ consists of null vectors, is totally isotropic by polarization, and has dimension at most $\lfloor k/2\rfloor$. As $\operatorname{Im}B(a)^T=L^\perp\supset L$, the image of $B(a)B(a)^T$ has dimension $\dim L^\perp-\dim L=2s(a)-k$. \end{proof} \subsection{The local pole budget} Put $F_0=M_X^T\mathcal B(X,Y)M_X\Lambda^{-1}$, linear in $Y$, and $\Omega=(F-F_0)\Lambda$, a skew-symmetric matrix, homogeneous of degree one. Since $\Omega$ is skew and $\Lambda$ symmetric, $P\Omega=-P\Omega^T=-P\Lambda F^T+P\Lambda F_0^T$, which is polynomial by \eqref{eq:cp-cross}, as $P$ and $F_0$ are linear; and, as $F=F_0+\Omega H$ and $H\Lambda=I$, the first identity in \eqref{eq:cp-diagonal} makes $\Omega H\Omega^T+\Omega F_0^T+F_0\Omega^T=F\Lambda F^T-F_0\Lambda F_0^T$ polynomial; so the negative-order terms of $\Omega H\Omega^T$ are cancelled only by terms whose pole order is at most that of $\Omega$. \begin{lemma}\label{lem:cp-dvr-budget} Let $f$ be an irreducible polynomial, real or complex, defining a pole of $\Omega$ of order $d>0$. Every maximal $\nu$-minor of $P$ is divisible by $f^{2d}$; if $d=1$ and the residue has rank $r$, every such minor is divisible by $f^r$. Let the \emph{reduced common denominator} $D$ of $\Omega$ be the least common multiple of the denominators of its entries, each entry written in lowest terms; a pole of order $d$ along $f$ contributes $f^d$, so $D$ need not be squarefree. Then $D^2$ divides every maximal minor of $P$, and $2\deg D\le\nu$. \end{lemma} \begin{proof} Localize the polynomial ring at $(f)$, a discrete valuation ring with uniformizer $\pi=f$. An alternating matrix over its fraction field is unimodularly congruent to two-by-two blocks $a_j\left[\begin{smallmatrix}0&1\\-1&0\end{smallmatrix}\right]$ and a zero block (pivot on an entry of least valuation and clear its two rows and columns; all coefficients are integral). Let the negative block valuations be $-d_j$. As $P\Omega$ is integral, the corresponding two columns of the transformed $P$ are divisible by $\pi^{d_j}$, so every maximal minor has valuation at least $2\sum_jd_j\ge2d$, since the largest $d_j$ is the pole order $d$. At a simple pole the residue rank is twice the number of negative blocks. Unimodular column changes preserve the maximal-minor ideal, and distinct irreducible factors multiply, which gives the claim for $D$. Some maximal minor is nonzero, of degree $\nu$, since $P$ is generically injective. \end{proof} If $\Omega$ has a pole of order $d>0$ along an irreducible $f$, call $W=(f^d\Omega)\bmod f$, a matrix over the residue field of $(f)$, the \emph{leading Laurent coefficient} of $\Omega$ at $f$. The leading terms of the same identities give, over the residue field, \begin{equation}\label{eq:cp-residue} PW=0,\qquad W^T=-W,\qquad WHW^T=0; \end{equation} the last because the order-$2d$ term of the upper-block identity has no other possible cancellation. So the rank of $W$ is even and its image is $H$-isotropic. For $\nu=8$ the budget reads $D^2\mid\det P$ and $\deg D\le4$.