\begin{table}[H]
\centering
\begin{threeparttable}
\caption{Minimum route-screen membership stability for $k=2,3,4,5$}
\label{tab:topksensitivity}
\small
\setlength{\tabcolsep}{4pt}
\begin{tabularx}{0.96\textwidth}{>{\raggedright\arraybackslash}p{0.22\textwidth}>{\raggedright\arraybackslash}p{0.25\textwidth}>{\raggedright\arraybackslash}X>{\raggedright\arraybackslash}p{0.24\textwidth}}
\toprule
Sensitivity layer & Screen & Minimum overlap at $k=2,3,4,5$ & Minimum Jaccard at $k=2,3,4,5$ \\
\midrule
Geometry and jitter & Absolute facility density & 1/2, 1/3, 3/4, 4/5 & 0.33, 0.20, 0.60, 0.67 \\
Geometry and jitter & Relative facility screen & 1/2, 3/3, 3/4, 4/5 & 0.33, 1.00, 0.60, 0.67 \\
POI definition & Absolute facility density & 1/2, 2/3, 3/4, 5/5 & 0.33, 0.50, 0.60, 1.00 \\
POI definition & Relative facility screen & 1/2, 2/3, 3/4, 4/5 & 0.33, 0.50, 0.60, 0.67 \\
\bottomrule
\end{tabularx}
\begin{tablenotes}
\footnotesize
\item Notes: Geometry/measurement covers 48 buffer, pseudocount, and coordinate-jitter settings; POI definition covers seven category/denominator specifications. Each cell is the worst agreement with the 500 m baseline at the stated $k$. The exercise diagnoses proxy-ranking stability and is not a population estimate or causal result.
\end{tablenotes}
\end{threeparttable}
\end{table}