\section{Main theorems for positive determinant}\label{sec:main-thms} The next three statements isolate the positive-determinant case. Theorem~\ref{thm:all-infimum} later removes that determinant restriction, and Theorem~\ref{thm:finite-attainment-all} and Corollary~\ref{cor:all-safety} give the complete positive-data attainment and feasible two-hidden-state safety classification. The simpler case is treated first because it supplies the quantitative bias estimate. \begin{theorem}[Unrestricted first-moment infimum]\label{thm:H1} Under these assumptions, \begin{equation}\label{eq:H2} I_\infty(d)=I_4(d)=B(d). \end{equation} In particular, the conditions themselves guarantee a nonempty four-hidden-state class. One can approach the infimum while matching the \emph{same fixed} \(P,F,R,L\) exactly, using four hidden states, no hidden-to-hidden edges, and a uniform upper bound on all transition rates. The construction permits some rates to tend to zero. Four is a sufficient state budget, not a claimed minimum. If the two cross-excursion means differ, no finite compatible chain attains this unrestricted infimum. \end{theorem} \begin{theorem}[Complete two-hidden-state safety dichotomy]\label{thm:H2} Assume additionally \(\mathcal C_2(d)\ne\varnothing\). The reduced infimum is attained, and \begin{equation}\label{eq:H3} I_2(d)=I_\infty(d) \quad\Longleftrightarrow\quad \frac{L_{12}}{R_{12}}=\frac{L_{21}}{R_{21}}. \end{equation} Thus equality of the two conditional cross-excursion means is necessary and sufficient for safety of the \emph{infimum} within this regime. It is not a test that every individual fit has the same dissipation, nor a test identifying the true hidden network. Feasibility of the two-hidden-state class is an explicit hypothesis, not inferred from the single equality in \eqref{eq:H3}. \end{theorem} \begin{theorem}[Quantitative reduction bias]\label{thm:H3} Define \[\Delta m=\frac{L_{12}}{R_{12}}-\frac{L_{21}}{R_{21}},\qquad a=\operatorname{tr}(R^{-1}L),\qquad b=\frac{\det L}{\det R}.\] Under the hypotheses of Theorem~\ref{thm:H2}, \(a,b>0\) and \begin{equation}\label{eq:H4} I_2(d)-I_\infty(d)\ge \frac{2(R_{12}+R_{21})(\Delta m)^2}{(a+2\sqrt b)^2}. \end{equation} For \(\Delta m\ne0\) the gap is strictly positive and applies to every exact two-hidden-state fit. This bound is not asserted to be the optimal gap constant. \end{theorem}