spb/firc0001_uqo

1% Generated by code/make_tables.py from results/focus/focus\_coverage\_by\_form\_item.csv, focus\_coverage\_by\_form.csv, stats\_keys.json. Do not edit by hand.2\begin{table}[H]3\centering\scriptsize4\setlength{\tabcolsep}{2.5pt}5\caption{Focus-window coverage: share of each ledgered section's text actually shown to the model, by form and item}\label{tab:focus}6\begin{threeparttable}7\begin{adjustbox}{max width=\textwidth}8\begin{tabular}{llrrrrrrrr}9\toprule10Form & Item & \multicolumn{1}{c}{\makecell{Ledgered\\rows}} & \multicolumn{1}{c}{\makecell{Over 9{,}000\\chars (\%)}} & \multicolumn{1}{c}{\makecell{Median\\chars}} & \multicolumn{1}{c}{\makecell{Share shown\\P10}} & \multicolumn{1}{c}{\makecell{Share shown\\median}} & \multicolumn{1}{c}{\makecell{Share shown\\P90}} & \multicolumn{1}{c}{\makecell{Rows shown\\$<25$\% (\%)}} & \multicolumn{1}{c}{\makecell{Mentions from\\such rows (\%)}} \\11\midrule1210-K & \texttt{business} & 29{,}285 & 81.1 & 37{,}168 & 5 & 16 & 100 & 66.6 & 86.1 \\1310-K & \texttt{full\_body} & 1{,}230 & 99.9 & 304{,}265 & 1 & 2 & 4 & 99.8 & 100.0 \\1410-K & \texttt{md\_and\_a} & 28{,}919 & 94.0 & 64{,}194 & 4 & 9 & 24 & 90.5 & 96.7 \\1510-K & \texttt{properties} & 28{,}236 & 5.6 & 1{,}255 & 100 & 100 & 100 & 3.2 & 11.3 \\1610-K & \texttt{quantitative\_disclosures} & 26{,}876 & 11.9 & 3{,}354 & 45 & 100 & 100 & 5.4 & 33.2 \\1710-K & \texttt{risk\_factors} & 29{,}158 & 96.8 & 62{,}542 & 4 & 10 & 28 & 87.9 & 93.0 \\18\midrule1910-Q & \texttt{controls\_and\_procedures} & 92{,}393 & 1.9 & 1{,}710 & 100 & 100 & 100 & 1.1 & 3.2 \\2010-Q & \texttt{financial\_statements} & 91{,}882 & 96.8 & 64{,}457 & 5 & 11 & 22 & 92.4 & 94.2 \\2110-Q & \texttt{full\_document} & 241 & 100.0 & 348{,}536 & 1 & 2 & 7 & 100.0 & 100.0 \\2210-Q & \texttt{legal\_proceedings} & 66{,}998 & 3.3 & 522 & 100 & 100 & 100 & 1.7 & 10.9 \\2310-Q & \texttt{md\_and\_a} & 91{,}371 & 94.4 & 47{,}488 & 6 & 13 & 31 & 84.7 & 89.8 \\2410-Q & \texttt{quantitative\_disclosures} & 83{,}138 & 4.9 & 1{,}510 & 100 & 100 & 100 & 1.0 & 7.7 \\2510-Q & \texttt{risk\_factors} & 72{,}560 & 22.9 & 969 & 7 & 100 & 100 & 17.7 & 42.9 \\2610-Q & \texttt{unregistered\_sales} & 80{,}210 & 6.9 & 3{,}886 & 100 & 100 & 100 & 4.3 & 5.7 \\27\midrule288-K & \texttt{full\_body} & 400{,}601 & 8.8 & 3{,}816 & 100 & 100 & 100 & 3.3 & 5.7 \\29\midrule3010-K & all items & 143{,}704 & 59.3 & 28{,}255 & 5 & 22 & 100 & 52.1 & 84.0 \\3110-Q & all items & 578{,}793 & 35.5 & 3{,}503 & 8 & 100 & 100 & 31.4 & 79.2 \\328-K & all items & 400{,}601 & 8.8 & 3{,}816 & 100 & 100 & 100 & 3.3 & 5.7 \\33\bottomrule34\end{tabular}35\end{adjustbox}36\begin{tablenotes}\footnotesize37\item The snapshot's \texttt{\_focus} function was replayed on all 1{,}123{,}098 ledgered sections (the replicated spans equal the real output). Sections up to 9{,}000 characters are shown whole; longer sections are reduced to windows around candidate-gate hits up to the budget, so the ``share shown'' is the fraction of the section's characters contained in the text sent to the model. The last column is the share of the item's ledger mentions that come from sections of which less than a quarter was shown (193{,}416 of 308{,}477 ledger mentions overall). No user message reached the 14{,}000-character cut. Shares are over all ledgered rows, including gate-rejected ones (calls-only shares in \texttt{results/focus/focus\_coverage\_calls\_only\_by\_form\_item.csv}). Coverage of the text is not recall: what the model missed inside the shown windows is unmeasured.38\end{tablenotes}39\end{threeparttable}40\end{table}41