\begin{table}[H]
\centering
\begin{threeparttable}
\caption{POI-definition sensitivity of route-screen membership}
\label{tab:proxydefinitions}
\small
\setlength{\tabcolsep}{4pt}
\begin{tabularx}{0.96\textwidth}{>{\raggedright\arraybackslash}Xrrrr}
\toprule
Definition & Sensitive features & Service opportunities & Absolute & Relative \\
\midrule
All mapped features & 2,933 & 6,195 & 3/3 & 3/3 \\
Omit education/childcare & 695 & 6,195 & 2/3 & 3/3 \\
Omit health care & 2,256 & 6,195 & 2/3 & 3/3 \\
Omit eldercare/other & 2,915 & 6,195 & 3/3 & 3/3 \\
Food-service denominator & 2,933 & 3,652 & 3/3 & 3/3 \\
Retail/market denominator & 2,933 & 477 & 3/3 & 2/3 \\
Civic/recreation denominator & 2,933 & 2,066 & 3/3 & 2/3 \\
\bottomrule
\end{tabularx}
\begin{tablenotes}
\footnotesize
\item Notes: Counts describe the citywide processed OSM layers. Absolute and relative columns report overlap with the baseline top three. The three leave-one-family-out rows alter the sensitive-facility numerator; the final three rows alter the service-opportunity denominator. These are proxy-definition tests, not validation against observed deliveries or acoustic exposure.
\end{tablenotes}
\end{threeparttable}
\end{table}