1\newpage2\begin{appendices}3\section{Registered hypotheses}4\label{app:hypotheses}5\input{appendix/hypotheses_verbatim}67\section{Source audit and unexecuted methods}8\label{app:audit}9Thirty-four sources were verified on 2026-10-29 from their arXiv pages, repositories and documentation (\texttt{literature/sources.csv};10notes in \texttt{literature/audit\_group\{1,2,3\}.md}). Table~\ref{tab:audit} summarizes the core methods.11\begin{table}[H]\centering\scriptsize12\caption{Audited core methods: availability of code and checkpoints, backend, question access, and WP30 status.}\label{tab:audit}13\begin{tabular}{@{}llllll@{}}\toprule14Method & Code & Checkpoint & Backend & Question access & WP30 \\\midrule15SwaT (2610.15095) & yes (no licence) & none & torch/CUDA & blind & not run \\16Attention Matching (2602.16284) & MIT & n/a (optimization) & vLLM 0.11 / torch & blind & not run (MLX port out of scope) \\17C2C (2510.03215) & Apache-2.0 & Qwen3 fusers & torch/CUDA & blind & not run \\18LatentMAS (2511.20639) & Apache-2.0 & n/a & HF/vLLM, CUDA & blind & not run \\19KIVI (2402.02750) & MIT & n/a & CUDA kernels & n/a & motivation tested via B2 \\20SnapKV (2404.14469) & Apache-2.0 & n/a & torch & aware & faithful re-implementation (B5) \\21PyramidKV (2406.02069) & MIT & n/a & torch & aware & budget rule re-implemented (B6) \\22H$_2$O (2306.14048) & no LICENSE file & n/a & torch & cumulative & not run \\23StreamingLLM (2309.17453) & MIT & n/a & torch & blind & retention pattern only (B3) \\24LLMLingua-2 (2403.12968) & MIT & yes (HF) & torch (MPS ok) & blind & official (A8) \\\bottomrule25\end{tabular}26\end{table}2728\section{Correctness tests}29\label{app:tests}30\texttt{tests/test\_cache\_correctness.py} (12 tests, all passing on Qwen2.5-1.5B-Instruct-bf16): prefix equivalence in float32 and bf1631(distributional tolerance of Section~\ref{sec:correctness}, identical greedy continuation); no duplicated boundary token; identity32round trip with exact logits and measured bytes equal to \eqref{eq:mkv}; loader rejection of a wrong prefix hash and of a different33checkpoint; 100\,\% retention exact (shared and per-head indices); sparse mask path versus token-by-token append; preserved versus34relabelled positions differ; packed quantization bit widths and execution for 8/4/2 bits; no shared mutable state across branches;35attention selection keeps the window; no-message control. \texttt{tests/test\_leakage.py} (3) and \texttt{tests/test\_scoring\_and\_data.py} (5)36cover evaluator isolation, scorer behaviour and generator determinism.3738\section{Prompts}39\label{app:prompts}40\input{appendix/prompts_verbatim}4142\section{Pseudocode}43\label{app:pseudo}44\begin{algorithm}[h]\caption{Trial scheduler (resumable, append-only)}\footnotesize45\begin{algorithmic}[1]46\State load config, manifest, resources; $done \gets$ trial keys with status success in \texttt{runs/raw/<exp>.jsonl}47\For{document $d$ in selected items} \For{model pair $(S,R)$} \For{condition $c$} \For{seed}48 \State $P \gets$ questions of $d$ whose trial key $\notin done$; \textbf{continue} if $P=\emptyset$49 \If{$c$ is question-blind} build message/cache $m$ once for $d$ (reuse count $=|P|$) \EndIf50 \For{question $q \in P$} build $m$ if question-aware; run receiver; score with evaluator-only key; append validated record \EndFor51\EndFor\EndFor\EndFor\EndFor52\end{algorithmic}\end{algorithm}53\begin{algorithm}[h]\caption{Cache handoff pipeline (Track B/C)}\footnotesize54\begin{algorithmic}[1]55\State $\text{pre} \gets$ chat tokens of (system, document) that are a true prefix of the full (system, document, question) prompt56\State $C \gets$ receiver.prefill(pre) \Comment{dense bf16; measured doc prefill}57\State $C' \gets$ compress($C$): quantize / select positions (sparse cache with logical offset $|\text{pre}|$) / map from sender58\State payload $\gets$ serialize($C'$, manifest); optionally codec59\State at receiver: check manifest; deserialize; install; prefill fresh query tokens with retained-positions mask; decode greedily60\end{algorithmic}\end{algorithm}61\begin{algorithm}[h]\caption{Evidence retrieval (A6) and repairable chains}\footnotesize62\begin{algorithmic}[1]63\State refs $\gets$ A5 facts without values; rank refs by BM25(question, entity+attribute)64\For{ref in ranked order while lookups $<$ max and bytes $<$ budget} fetch source span $\pm$ context; count bytes and lookups \EndFor65\State chains: for codes mentioned in the previous message, fetch source sentences containing the code under the same meters66\end{algorithmic}\end{algorithm}67\begin{algorithm}[h]\caption{Adaptive policy (E8)}\footnotesize68\begin{algorithmic}[1]69\State on dev: for each (family, context bin, pair, access) and action $a$: $\hat q_a$, lower bootstrap bound $\underline{q}_a$, mean cost $\hat c_a$70\State feasible $\gets \{a : \underline{q}_a \ge \tau\,\max_{a'}\underline{q}_{a'}\}$; choose $\arg\min_{a\in\text{feasible}}\hat c_a$; fallback to the dev-best static action71\State freeze; on test: look up the cell (nearest bin fallback counted), apply the chosen action, add router time72\end{algorithmic}\end{algorithm}7374\section{Per-attempt record schema}75\label{app:schema}76Each attempt is one JSON line validated against \texttt{wp30-attempt-1}: identifiers (study, experiment, trial, attempt, run group,77protocol and config hashes, git commit, timestamp), task (family, document id and hash, question id, split, cluster), models (sender and78receiver revisions, tokenizer hashes, backend versions, hardware id, device state, weight precision), method (id, fidelity, compression79config with realized values, question access, topology, handoff index, reuse count, cache hit, compatibility hash, seed), sizes (source80tokens under both tokenizers, query and message tokens, UTF-8 bytes, cache tensor bytes, serialized payload bytes, side information,81fetched bytes, transport bytes), timing spans and TTFT, usage and provider cached usage, memory (process RSS, MLX active/peak, prefill and82decode peaks, mapper memory), cost with source, prediction and evidence artifacts, key reference, scores, status, error code, retry link,83log path, payload and output hashes.8485\section{Additional tables}86\label{app:tables}87\input{appendix/extra_tables}88\end{appendices}89