\subsection{Core results by question type}
\begin{table}[H]\centering\scriptsize
\caption{Synthetic family, core run, accuracy (\%) by question type (both pairs pooled; 95\,\% cluster-bootstrap CI).}
\adjustbox{max width=\textwidth}{\input{tables/table_qtype_core}}
\end{table}

\subsection{Attempt outcomes by method}
\begin{table}[H]\centering\scriptsize
\caption{All attempts of the core run by status (retries included).}
\input{tables/table_failures}
\end{table}

\subsection{Exploratory: question types not covered by the core (A-004)}
\label{app:qtypes}
The five-question core limit kept only single-fact, exact-copy and multi-hop questions. This exploratory run (5\,600 trials; amendment A-004;
no confirmatory test) asks the remaining ten question types of the 40 synthetic test documents for seven conditions. Three observations, all
exploratory: (i) \emph{revision/precedence} questions (``current budget under the later-date rule'') are the ones that lossy text loses most---full
text 75--88\,\%, blind summaries 12--20\,\%, structured evidence 4--8\,\%, sink+recent 10--15\,\%---while the question-aware attention cache keeps
76\,\% on Llama but only 28\,\% on Qwen; (ii) \emph{negation} (cancelled vs postponed, yes/no) is answered at 72--96\,\% by every condition that
retains anything, close to the base rate of the majority answer, and must not be read as evidence of fact transfer; (iii) \emph{aggregation}
(summing five to six amounts) is at 0\,\% for every condition including full text, and \emph{unit conversion} is solved only by Qwen (57--60\,\%
with full text, 8-bit cache or aware eviction) and never from a summary---these are receiver-capability limits, not communication effects.
Unanswerable questions were abstained on correctly 94--100\,\% of the time in every condition.
\begin{figure}[H]\centering\includegraphics[width=\textwidth]{figA_qtypes_exploratory.pdf}
\caption{EXPLORATORY. Accuracy on negation, revision/precedence, temporal order, aggregation, unit conversion and unanswerable questions
(questions 6--15 of the 40 synthetic test documents) for a subset of methods; no confirmatory test attached.}\end{figure}
\begin{table}[H]\centering\scriptsize
\caption{EXPLORATORY question-type extension (\%, 95\,\% CI).}
\adjustbox{max width=\textwidth}{\input{tables/table_qtypes_exploratory}}
\end{table}

\subsection{Question-access panels}
\begin{figure}[H]\centering\includegraphics[width=\textwidth]{figA_quality_vs_bytes_blind.pdf}
\caption{Question-blind conditions only (reusable messages).}\end{figure}
\begin{figure}[H]\centering\includegraphics[width=\textwidth]{figA_quality_vs_bytes_aware.pdf}
\caption{Question-aware conditions only (query-specific messages).}\end{figure}