\subsection{All-population information bounds} The following bounds hold at every population size. They use the classical conditional-score principle and a model-specific quadratic projection, together with the prediction-risk form of missing information. Define \begin{align} H_0(t)&=\frac{4t}{e^{4t}-1},& B_n(t)&=\frac{4t(e^{4t}+1)} {(e^{4t}-1)\{n(e^{4t}-1)+2\}},\label{eq:finite-endpoint-bounds}\\ c_n&=1+\frac1n,& \ell_n(\lambda)&=\frac{c_n\lambda e^{-\lambda}}{1+\lambda^2/2},\notag\\ U_n(\lambda)&=1+\lambda- \left[1+\left(1-\frac1n\right)\lambda\right]e^{-\lambda}. &&\label{eq:finite-loss-bounds} \end{align} Both $H_0$ and $B_n$ have continuous value one at zero, while $\ell_n(0)$ and $U_n(0)$ are zero. Whenever a zero-time information fraction is used, it means the limit $R_n(0)=1$, not the quotient in \eqref{eq:R} at zero. \begin{theorem}[All-population information bounds]\label{thm:finite-certificate} In the stationary experiment of Proposition~\ref{prop:exact}, for every integer $n\ge1$ and $t>0$, \begin{equation}\label{eq:finite-certificate} \max\{B_n(t),1-U_n(nt)\}\le R_n(nt) \le\min\{H_0(t),1-\ell_n(nt)\}. \end{equation} In particular, \begin{equation}\label{eq:finite-loss-certificate} \ell_n(\lambda)\le1-R_n(\lambda)\le U_n(\lambda), \qquad\lambda>0. \end{equation} For $n=1$, the labelled and aggregate endpoint experiments coincide, and $R_1(t)=B_1(t)=H_0(t)$. \end{theorem} The proof of Theorem~\ref{thm:finite-certificate} is in Appendix~\ref{app:finite-proofs}. \paragraph{Relation to the sharp window.} The quadratic lower bound has a direct connection to the matched Gaussian information $g(t)$ in \eqref{eq:g}: \begin{equation}\label{eq:finite-gaussian-identity} nt B_n(t)=g(t)\frac{n(e^{4t}-1)}{n(e^{4t}-1)+2}. \end{equation} It has the correct leading information when $\lambda\to\infty$ and $\lambda=o(n)$, since $\lambda B_n(\lambda/n)\to1/2$, and at fixed $00$, $B_n(\lambda/n)\to1/(1+2\lambda)$; the lower bound does not recover the whole function $F(\lambda)$. For $n\ge2$, the quadratic projection defined above satisfies \begin{equation}\label{eq:finite-projection-residual} \lambda\{R_n(\lambda)-B_n(\lambda/n)\} =\E[(S-\Pi_2S)^2]. \end{equation} Section~\ref{sec:finite-residual} gives an explicit upper certificate for this residual. It encloses information relative to the quadratic reference $B_n$. Together with the clipped-score coupling, it also proves the sharp relative comparison with $F$ in Section~\ref{sec:window}.