% Auto-generated by scripts/10_stage_e_figures.py — do not edit numbers.
\begin{table}[t]\centering\begin{adjustbox}{max width=\textwidth}
\begin{threeparttable}
\caption{Sample construction by decade}\label{tab:sample_construction}
\small\begin{tabular}{lrrrr}
\toprule
 & 1992–1999 & 2000–2009 & 2010–2019 & 2020–2026 \\
\midrule
Raw bar rows & 626,407,882 & 2,331,426,585 & 2,657,373,989 & 2,071,067,068 \\
Duplicate rows merged & 0 & 0 & 0 & 10,128 \\
OHLC-violation flags & 90 & 10 & 0 & 0 \\
Split-window flagged rows & 772,707 & 1,811,211 & 815,630 & 897,786 \\
Regular-hours bar rows & 611,399,568 & 2,278,758,476 & 2,599,857,420 & 1,940,283,624 \\
Distinct symbols & 26,519 & 22,052 & 18,761 & 46,985 \\
Trading days & 2,022 & 2,515 & 2,516 & 1,569 \\
\bottomrule\end{tabular}
\begin{tablenotes}[flushleft]\footnotesize
\item Notes: Full panel (all instruments). Duplicates merged by the rule
open=first, high=max, low=min, close=last, volume=sum. Flags mark rows;
no raw data are deleted. Split windows are $\pm 1$ trading day around
high-confidence overnight-ratio events (Section~\ref{sec:data}).
\end{tablenotes}\end{threeparttable}
\end{adjustbox}\end{table}