\section{Holding-time information loss}\label{sec:holding} Constant escape is essential to replacing the complete score by a jump count. Let $Q$ be an irreducible finite-state row generator on at least two states, with off-diagonal rates $q_{ij}$, escape rates $a_i=\sum_{j\ne i}q_{ij}$ and stationary law $\pi$. Take $X_0\sim\pi$; this initial law is independent of the common clock parameter. The unknown parameter multiplies $Q$. In dimensionless time $t=\theta\Delta$, the complete log-$\theta$ score is \begin{equation}\label{eq:holdingscore} S_t=K_t-\int_0^t a(X_u)\,du, \qquad \E S_t^2=\bar a t,\quad \bar a=\sum_i\pi_i a_i. \end{equation} The endpoint-conditioned count and dwelling-time machinery, including the observed-score and missing-information identities, is established \cite{HobolthJensen2005,HobolthJensen2011}; the following coefficient makes the holding-time contribution to information loss explicit. \begin{proposition}[Irreducible holding-time loss]\label{prop:holding} Let $\mathcal H$ contain the initial state and the ordered visited-state sequence, but not the jump times. As $t\downarrow0$, \begin{equation}\label{eq:holdinglower} \E\Var(S_t\mid\mathcal H) =\frac{t^3}{12}\sum_{i,j\ne i}\pi_iq_{ij}(a_i-a_j)^2+O_Q(t^4). \end{equation} The missing information when only the endpoints are observed is at least this quantity. \end{proposition} \begin{proof} For exactly one jump $i\to j$, its time $U$ has conditional density proportional to $e^{-a_i u-a_j(t-u)}$ on $[0,t]$. Rescaling by $t$ and expanding the bounded exponential shows $\Var(U\mid\mathcal H)=t^2/12+O_Q(t^3)$. On this event, $S_t=1-a_jt-(a_i-a_j)U$, and the event probability is $\pi_iq_{ij}t+O_Q(t^2)$. Summation gives the displayed leading term. For no jumps there is no conditional variance. On a skeleton with at least two jumps, the integral in \eqref{eq:holdingscore} lies in an interval of length at most $a_{\max}t$, whereas $K_t$ is fixed by the skeleton. Such skeletons have total probability $O_Q(t^2)$ by Poisson domination, giving an $O_Q(t^4)$ contribution. Finally, the endpoint observation is a function of $\mathcal H$, so conditional variance decomposition proves the lower bound. \end{proof} \subsection{A local classification for every finite irreducible chain} The holding-time term can be combined with path aliases to classify the first loss order for every finite irreducible chain. Write $B$ for the off-diagonal part of $Q$, so that $B_{ii}=0$, and let $m_{ij}$ be the shortest directed-walk length from $i$ to $j$, with $m_{ii}=0$. Define the graph alias depth $h$ as the smallest second distinct walk length among all ordered endpoint pairs. Irreducibility ensures $h<\infty$. Put \begin{align} A_2(Q)&=\frac12\sum_i\pi_i\left[ 4(B^2)_{ii}+\sum_{j\ne i:\,q_{ij}>0}(B^2)_{ij}\right], \label{eq:A2}\\ H(Q)&=\frac1{12}\sum_{i,j\ne i}\pi_iq_{ij}(a_i-a_j)^2. \label{eq:Hgeneral} \end{align} When $A_2(Q)=0$, also define \begin{equation}\label{eq:A3} A_3(Q)=\frac16\sum_{i,j:\,m_{ij}<3} \pi_i(3-m_{ij})^2(B^3)_{ij}. \end{equation} \begin{theorem}[Endpoint information at small times]\label{thm:classification} For a fixed finite irreducible $Q$ on at least two states, let \begin{equation}\label{eq:generalM} \mathcal M_Q(t)=\bar a t- t^2\sum_{i,j}\pi_i\frac{((Qe^{tQ})_{ij})^2}{(e^{tQ})_{ij}}. \end{equation} This is the missing log-clock information between the full trajectory and its endpoints. With $A_2$, $H$, $A_3$ from \eqref{eq:A2}--\eqref{eq:A3} and the graph alias depth $h$ defined above, as $t\downarrow0$, \begin{equation}\label{eq:Mtwo} \mathcal M_Q(t)=A_2(Q)t^2+O_Q(t^3). \end{equation} If $A_2(Q)=0$, then \begin{equation}\label{eq:Mthree} \mathcal M_Q(t)=[A_3(Q)+H(Q)]t^3+O_Q(t^4). \end{equation} Consequently, if escape rates are not constant, the fractional information loss has power $\min\{h-1,2\}$. If they are constant, its power is $h-1$ with the coefficient given by the graph form of Theorem~\ref{thm:alias}. All remainders concern a fixed generator, not a family with vanishing edges or increasing state space. \end{theorem} The proof of Theorem~\ref{thm:classification} is in Appendix~\ref{app:fixed-generator-details}. For a heterogeneous three-cycle with rates $(1,2,3)$, for example, $\bar a=18/11$, $A_2=0$, $A_3=9$, and $H=3/11$. Hence $\mathcal M_Q(t)=(102/11)t^3+O(t^4)$: count and holding-time effects enter at the same order and add. For the Ehrenfest chain with $\Omega$ particles, the same formula gives $A_2=\Omega(\Omega+1)$ and $H=0$. Dividing by the full information $\Omega t$, the leading loss of the retained fraction is $(1+1/\Omega)\lambda$ with $\lambda=\Omega t$. \begin{remark}[Reversible and birth--death generators]\label{rem:birth-death} If $Q$ is reversible, every allowed edge $i\to j$ has $q_{ji}=\pi_iq_{ij}/\pi_j>0$, so $(B^2)_{ii}\ge q_{ij}q_{ji}>0$ and $A_2(Q)>0$. The endpoint loss of a reversible generator is therefore always of order $t^2$; the $t^3$ regime of \eqref{eq:Mthree}, where $H$ enters, requires a generator without two-cycles. For a birth--death chain with birth rates $\beta_i$, death rates $\delta_i$ (absent rates set to zero) and $a_i=\beta_i+\delta_i$, there are no two-step shortcuts, and detailed balance $\pi_i\beta_i=\pi_{i+1}\delta_{i+1}$ gives \[ A_2(Q)=2\sum_i\pi_i(\beta_i\delta_{i+1}+\delta_i\beta_{i-1}),\qquad H(Q)=\frac16\sum_i\pi_i\beta_i(a_{i+1}-a_i)^2 . \] Here $H$ is the explicit holding-time lower bound of Proposition~\ref{prop:holding}, not a leading-order term. For the Ehrenfest chain, $a_i\equiv\Omega$ gives $H=0$ and $A_2=\Omega(\Omega+1)$, as above. \end{remark}