\section{Finite information certificates}\label{app:finite-proofs} \subsection{All-population endpoint bounds} The following is the complete proof of Theorem~\ref{thm:finite-certificate}. \begin{proof} The complete log-clock score is $C=K-\lambda$, where $K\sim\Pois(\lambda)$. For $Y=(N_0,N_\Delta)$, its conditional expectation $S=\E[C\mid Y]$ is the endpoint log-clock score. Hence \begin{equation}\label{eq:finite-score-risk} \E S^2=\lambda R_n(\lambda),\qquad \lambda(1-R_n(\lambda))=\inf_f\E[(K-f(Y))^2]. \end{equation} The infimum is over square-integrable endpoint predictors. \paragraph{Quadratic projection.} Put $p=(1-e^{-2t})/2$, $u=n(1-p)$, $q=np$, and \[ A=N_0+N_\Delta-n,\qquad D=N_\Delta-N_0. \] In the independent stationary-particle representation, each particle contributes $(\pm1,0)$ with probabilities $(1-p)/2$, or $(0,\pm1)$ with probabilities $p/2$. The $n$ particle contributions are independent; successive observation intervals are not. Their second and fourth moments give \[ \E A^2=u,\quad \E D^2=q,\quad \E A^4=3u^2+u(-2+3p),\quad \E D^4=3q^2+q(1-3p), \] and $\E A^2D^2=n(n-1)p(1-p)$. For example, the last identity follows by expanding the sums and using that a single particle cannot contribute to both axes. Therefore the centered quadratics $W_1=A^2-u$ and $W_2=D^2-q$ have covariance matrix \begin{equation}\label{eq:finite-quadratic-covariance} \Gamma=\begin{pmatrix} 2u^2+u(-2+3p)&-np(1-p)\\ -np(1-p)&2q^2+q(1-3p) \end{pmatrix}. \end{equation} Freeze the centering constants $u,q$ at the parameter value being evaluated. The score identity differentiates the raw moments, at fixed physical spacing $\Delta$, and gives \begin{equation}\label{eq:finite-quadratic-score} v:=\begin{pmatrix}\E[SW_1]\\\E[SW_2]\end{pmatrix} =nt e^{-2t}\begin{pmatrix}-1\\1\end{pmatrix}. \end{equation} Here the log-clock derivative is $t\partial_t$, and $p'(t)=e^{-2t}$. For $n\ge2$, \[ \det\Gamma=2n^2p(1-p)(n-1) \{1+2(n-2)p(1-p)\}>0. \] The orthogonal projection of $S$ onto $\operatorname{span}\{W_1,W_2\}$ has squared norm $v^\top\Gamma^{-1}v$. Direct inversion gives \begin{equation}\label{eq:finite-projection-bound} \frac{v^\top\Gamma^{-1}v}{nt} =H_0(t)\frac{(1-p)^2+p^2}{1+2(n-2)p(1-p)}=B_n(t). \end{equation} Thus $R_n(nt)\ge B_n(t)$. For $n=1$, this covariance matrix is singular; the direct parity calculation \eqref{eq:omegaone} instead yields $R_1(t)=H_0(t)=B_1(t)$. \paragraph{Labelled endpoints.} Revealing every particle's initial and final state refines the aggregate observation. Each parity-flip indicator is Bernoulli with probability $p$, and its log-clock information is $(tp'(t))^2/[p(1-p)]$. The labelled experiment therefore has information $n(tp'(t))^2/[p(1-p)]$. Dividing by $nt$ and applying data processing gives \[ R_n(nt)\le\frac{t e^{-4t}}{p(1-p)}=H_0(t). \] These indicators count particles with different endpoint states, not the actual number of reactions $K$. \paragraph{Prediction by the endpoint displacement.} The predictor $f(Y)=|D|$ recovers $K$ when $K\le1$. At $K=2$ its squared error is four exactly on a return to the initial aggregate count. The embedded chain selects a uniformly random particle at each event. With probability $1/n$ the same particle is selected twice; otherwise a return requires opposite initial states, with probability $1/2$. Thus the stationary two-step return probability is \begin{equation}\label{eq:finite-return} \bar r_n=\frac1n+\frac12\left(1-\frac1n\right)=\frac{c_n}{2}. \end{equation} For $K\ge3$, $|D|\le K$ gives $(K-|D|)^2\le K^2$. Consequently \[ \lambda(1-R_n(\lambda))\le c_n\lambda^2e^{-\lambda}+\E[K^2\one_{\{K\ge3\}}]. \] Since $\E K^2=\lambda+\lambda^2$, removal of the $K=1,2$ terms gives \[ \frac{\E[K^2\one_{\{K\ge3\}}]}{\lambda} =1+\lambda-(1+2\lambda)e^{-\lambda}. \] This proves the upper loss bound $1-R_n\le U_n$. \paragraph{Retaining only zero jumps and two-jump returns.} Deleting other histories decreases every squared prediction risk, and therefore decreases its infimum in \eqref{eq:finite-score-risk}. Given $N_0=i$, write $r_i=[2i(n-i)+n]/n^2$ for the embedded two-step return probability. The retained diagonal weights are \[ w_0=e^{-\lambda},\qquad w_2=e^{-\lambda}\lambda^2r_i/2. \] The minimum unnormalized squared risk for these two masses is \[ \frac{4w_0w_2}{w_0+w_2} =\frac{2\lambda^2e^{-\lambda}r_i}{1+\lambda^2r_i/2}. \] Average over the stationary initial count. Since $0\le r_i\le1$ and $\E r_i=\bar r_n=c_n/2$, bounding the denominator above by $1+\lambda^2/2$ yields \[ \lambda(1-R_n(\lambda))\ge \frac{c_n\lambda^2e^{-\lambda}}{1+\lambda^2/2}. \] This proves the lower loss bound and completes \eqref{eq:finite-certificate}. After cancellation of the common Poisson weight, the rational denominator $1+\lambda^2r_i/2$ is at least one. No rare endpoint probability needs to be bounded below in this argument. \end{proof}