1\section{Instrument provenance and data lineage}\label{app:provenance}23\paragraph{Lineage.} Figure~\ref{fig:lineage} summarizes the audited pipeline: FinData OS export jobs (dataset identifier,4version, SHA-256 per file, recorded in \texttt{results/manifests/ingest\_manifest.json}) $\to$ daily panels per instrument5(returns, overnight split, realized measures, validity flags; hashes in \texttt{panels\_manifest.json}) $\to$ rolling6forecasts (one Parquet per instrument and specification; run manifest with git SHA, host, package versions, wall and CPU7time, peak memory) $\to$ per-origin losses $\to$ analysis outputs $\to$ exhibits (sources listed in8\texttt{figure\_audit.csv}). The pre-registration tag \texttt{prereg-frozen} precedes every test-period loss.910\paragraph{Datasets.} ETFs and stocks: \texttt{market\_bars.\{etf,stock\}.1m.split} (one-minute, split-adjusted) for realized11measures and \texttt{...1d.split\_div} (daily, split- and dividend-adjusted) for returns; futures:12\texttt{market\_bars.futures.1m.continuous\_ratio} (vendor roll dates, ratio back-adjustment); FX: \texttt{market\_bars.fx.1m.raw}13(mid quotes from four money-center banks, no daily dataset exists so daily returns are built from the last bar of the1417:00--17:00 ET day); crypto: \texttt{market\_bars.crypto.1m.raw} (UTC days). Index levels (VIX, SPX) from15\texttt{market\_bars.index.1d.raw} are used descriptively.1617\paragraph{Audit evidence (2026-11-09).} 41 OpenAPI paths; bars schema v1 with bar-open UTC timestamps; AAPL raw/split close ratio184.000 on 2020-08-28 and 1.000 on 2020-08-31 (4:1 split); SPY split/split-dividend close ratio 1.0327 before and 1.0295 after the192024-03-15 ex-dividend date; ES ratio-adjusted/unadjusted close ratio 1.1123 before and 1.0986 after the March 2024 roll, equal20to 1.000 on the last sample days; all daily rows carry \texttt{revision\_id} 1 and the vendor's \texttt{off\_interval} flag only;21SPY daily sessions match the XNYS calendar; export smoke test returned 252 rows for SPY 2024 with a recorded dataset version.2223\paragraph{Licence.} FirstRate Data: private use by the subscriber; no redistribution of bars; derived results publishable24with attribution. The replication package therefore excludes \texttt{data/raw}, \texttt{data/interim}, the panels and the25per-origin forecasts of returns, and includes code, configurations, manifests, checksums, losses, analysis outputs and a26synthetic fixture that exercises every stage.2728\paragraph{Literature search.} Crossref, arXiv, OpenAlex and web searches up to 2026-11-09; every reference in the bibliography29was verified against its DOI or a fetched publisher/preprint page (\texttt{results/metrics/references\_verification.json}); one30candidate (Chung 1999, working paper) could not be verified and is not cited.3132\begin{figure}[tp]\centering33\includegraphics[width=\textwidth]{figures/fig_lineage.pdf}34\caption{Reproducibility flow and audited data lineage (generated from the manifests).}\label{fig:lineage}35\end{figure}36