\subsection{Posterior controls for the finite Gaussian envelope} \label{app:finite-gaussian} The following is the complete proof of Theorem~\ref{thm:finite-gaussian}. Take $s,r,w,h,\nu,d,\gamma$ from \eqref{eq:residual-parameters}. First suppose $n\ge2$ and $t>0$ is finite, so $0
0$.
Let $M\sim\Bin(n,p)$ be the parity-change particle count, and put
$X=M-np$ and $H_2=X^2-rX-\nu$. This $M$ is not the actual reaction
count $K$. Write $\widehat H_2=\E[H_2\mid A,D]$.
\begin{proof}
\par\smallskip\noindent\textbf{A posterior recurrence with no boundary remainder.}\quad
At an admissible endpoint $(A,D)=(a,z)$, the posterior weights $q_m$ are
proportional to
\[
\frac{p^m s^{n-m}}
{\left(\frac{n-m+a}{2}\right)!
\left(\frac{n-m-a}{2}\right)!
\left(\frac{m+z}{2}\right)!
\left(\frac{m-z}{2}\right)!}.
\]
The complete feasible parity class has $m\ge|z|$ and $n-m\ge|a|$.
Its consecutive terms obey
\[
p^2\{(n-m)^2-a^2\}q_m
=s^2\{(m+2)^2-z^2\}q_{m+2}.
\]
Define the random polynomials
\[
\alpha=p^2\{(n-M)^2-A^2\},\qquad
\beta=s^2(M^2-D^2).
\]
At the highest posterior support point $\alpha=0$, and at the lowest
point $\beta=0$. Summing the recurrence gives, for every polynomial $f$,
\begin{equation}\label{eq:gaussian-control-identity}
\E[\alpha f(M+2)\mid A,D]=\E[\beta f(M)\mid A,D].
\end{equation}
This includes singleton supports. Hence
\[
T_f=(\alpha-\beta)f(M)+\alpha\{f(M+2)-f(M)\}
\]
has conditional mean zero. Use the three polynomials
\begin{equation}\label{eq:gaussian-controls}
f_0=1,\qquad f_1=X-1,\qquad f_2=H_2-2X+2+r,
\qquad T=(T_{f_0},T_{f_1},T_{f_2})^\top.
\end{equation}
Their coefficients are evaluated at the fixed parameter of interest.
\par\smallskip\noindent\textbf{The exact Gram calculation.}\quad
Put $\kappa=\E H_2^2=2n(n-1)w^2$, and define
$G=\E[TT^\top]/\kappa$ and $v=\E[H_2T]/\kappa$. Direct expansion gives
\begin{align}
v_0&=-r,\qquad v_1=-2nw+8w-1,
&v_2&=-r(4nw-10w+1),\label{eq:gaussian-cross-vector}\\
G_{00}&=2(nw-4w+1),
&G_{01}&=2r(2nw-6w+1),\notag\\
G_{02}&=-2(10nw^2-3nw-24w^2+10w-1),\notag\\
G_{11}&=2(2n^2w^2-27nw^2+5nw+54w^2-16w+1),\notag\\
G_{12}&=2r(7n^2w^2-57nw^2+8nw+90w^2-20w+1),\notag\\
G_{22}&=2(6n^3w^3-154n^2w^3+27n^2w^2+772nw^3\notag\\
&\hspace{27mm}-215nw^2+13nw-984w^3+334w^2-32w+1).
\label{eq:gaussian-gram}
\end{align}
The matrix is symmetric. For completeness, all entries follow by
substituting the controls in \eqref{eq:gaussian-controls}, conditioning
first on $M$, and using
\begin{align*}
\E[A^2\mid M]&=n-M,&\E[A^4\mid M]&=3(n-M)^2-2(n-M),\\
\E[D^2\mid M]&=M,&\E[D^4\mid M]&=3M^2-2M,\\
\E[A^2D^2\mid M]&=(n-M)M.
\end{align*}
The signs on the two axes are conditionally independent. The remaining
binomial moments, through degree eight, are obtained from
\[
\E M^j=\sum_{\ell=0}^j
\left\{\begin{matrix}j\\\ell\end{matrix}\right\}(n)_\ell p^\ell.
\]
This identity expands powers in the falling-factorial basis and counts
ordered distinct Bernoulli coordinates. As one entry check,
$T_{f_0}=L-dX-rH_2$ with the $L$ of the residual proof. Orthogonality
gives $\E[H_2T_{f_0}]=-r\kappa$, and the conditional fourth moments give
$\E T_{f_0}^2=2\kappa(nw+r^2)$, the displayed $v_0,G_{00}$.
\par\smallskip\noindent\textbf{A positive rational majorant for $n\ge4$.}\quad
Define
\begin{align}
Q(n,w)&=12n^3w^3-142n^2w^3+23n^2w^2
+624nw^3-240nw^2+24nw\notag\\
&\hspace{12mm}-1152w^3+784w^2-180w+14,
\label{eq:gaussian-Q}\\
P(n,w)&=-3w^2hn^2+w(76w^2-28w+4)n
+576w^3-536w^2+152w-14.
\label{eq:gaussian-P}
\end{align}
Expansion of the determinant gives
\[
\det G=8w^3(n-3)(n-2)^2 Q(n,w).
\]
For $k=n-4\ge0$ and $v_*=1/4-w\ge0$,
\begin{align*}
Q(n,w)&=12w^3k^3+(2w^3+23w^2)k^2
+w(64v_*^2+24v_*+14)k\\
&\quad+160v_*^3+72v_*^2+18v_*+5/2>0.
\end{align*}
Since $G$ is a Gram matrix, it is positive semidefinite; its nonzero
determinant makes it positive definite. Its inverse is therefore valid.
For every deterministic vector $c$, conditional Jensen and
$\E[T\mid A,D]=0$ give
\[
\E\widehat H_2^2\le\E(H_2-c^\top T)^2.
\]
Minimize the right side at $c=G^{-1}v$. Substitution of
\eqref{eq:gaussian-cross-vector}--\eqref{eq:gaussian-gram}, or the
adjugate identity, gives
\begin{equation}\label{eq:gaussian-norm-certificate}
\E\widehat H_2^2\le
C_n(p):=\kappa(1-v^\top G^{-1}v)
=\frac{n(n-1)w^2d\{3nwh+200w^2-110w+14\}}{Q(n,w)}.
\end{equation}
The exact gap is
\begin{equation}\label{eq:gaussian-norm-gap}
\frac{hd}{2}-C_n(p)=-\frac{d^2P(n,w)}{2Q(n,w)}.
\end{equation}
As a quadratic in $n$, $P$ has strictly negative leading coefficient
$-3w^2h$ and discriminant $-8w^2Z(w)$, where
\[
Z(w)=1006w^4-1940w^3+1086w^2-242w+19.
\]
On $0 0,\qquad n\ge2.
\]
Thus its projection is well-defined, with exact squared norm
$\lambda B_n(t)=\gamma^2\nu h/d$. The parity identity
$d\widehat H_1=L-r\widehat H_2$ and orthogonality of
$\widehat H_2$ to $W_1,W_2$ then give
\[
\lambda R_n(\lambda)=\gamma^2
\left\{\frac{\nu h}{d}+\frac{r^2}{d^2}\E\widehat H_2^2\right\}
<\gamma^2\left\{\frac{\nu h}{d}+\frac{r^2h}{2d}\right\}
=\frac{\gamma^2h}{2}=g(t).
\]
This is the exact endpoint experiment, without a Gaussian approximation
or a removed set of rare cells. For $n\ge4$, the more precise certificate
\eqref{eq:gaussian-norm-certificate} and its gap yield
\[
\lambda R_n(\lambda)
\le g(t)+\frac{\gamma^2r^2P(n,w)}{2Q(n,w)}