\begin{table*}[!ht]\centering\scriptsize
\setlength{\tabcolsep}{3pt}
\begin{tabular}{@{}rrrrrrrr@{}}
\toprule
disc radius $r$ (px) & area $\pi r^2$ (px$^2$) & $n$ & acc. & final at 1 & $\tau_{\mathrm{dec}}$ & $\tau_{\mathrm{conf}}$ & gap \\
\midrule
4 & 50 & 40 & 1.000 & 98\% & 1.05 {\scriptsize[1.00,1.15]} & 7.72 {\scriptsize[6.72,8.72]} & 6.67 {\scriptsize[5.65,7.68]} \\
8 & 201 & 40 & 1.000 & 95\% & 1.10 {\scriptsize[1.00,1.25]} & 6.00 {\scriptsize[5.45,6.68]} & 4.90 {\scriptsize[4.33,5.60]} \\
16 & 804 & 40 & 1.000 & 98\% & 1.05 {\scriptsize[1.00,1.15]} & 6.35 {\scriptsize[5.58,7.28]} & 5.30 {\scriptsize[4.50,6.23]} \\
32 & 3217 & 40 & 1.000 & 100\% & 1.00 {\scriptsize[1.00,1.00]} & 6.78 {\scriptsize[5.83,7.85]} & 5.78 {\scriptsize[4.83,6.85]} \\
64 & 12868 & 40 & 1.000 & 100\% & 1.00 {\scriptsize[1.00,1.00]} & 5.97 {\scriptsize[5.33,6.78]} & 4.97 {\scriptsize[4.33,5.78]} \\
\midrule
pooled & --- & 200 & 1.000 & 98\% & 1.04 {\scriptsize[1.01,1.08]} & 6.57 {\scriptsize[6.17,6.96]} & 5.53 {\scriptsize[5.13,5.92]} \\
\bottomrule
\end{tabular}
\caption{\textbf{The synthetic evidence-strength sweep} (Sec.~\ref{app:synth}). Accuracy is $1.000$ at every radius, so the manipulation failed and no slope is interpretable; the pooled gap at the evidence ceiling is the one observation that survives.}
\label{tab:supp_synth}
\end{table*}