1\begin{appendices}2\section{Source audit and compliance}\label{app:source}3The Phase~0 source audit (\texttt{reports/00\_source\_audit.md}) documents the three tables kept from the snapshot (\texttt{stores}, 4{,}082 rows; \texttt{products}, 465{,}388; \texttt{sync\_log}, 577{,}773), their fields, missingness and the platform's own compliance review per source: public catalogue endpoints only, polite rates, nothing behind authentication, retained attribution and removal on request. The research used no scraping of the platform and a single control request to its public statistics endpoint to date-stamp the counters.45\section{Data dictionary}\label{app:dictionary}6The full field-level dictionary of the cleaned tables (\texttt{listings\_clean}, \texttt{stores\_master}, \texttt{sibling\_stores}, \texttt{match\_pairs}, \texttt{match\_groups}, \texttt{sync\_log}) is generated by the pipeline (\texttt{reports/data\_dictionary.md}). Key derived fields: \texttt{title\_norm} (ASCII-folded, lower-cased title used for duplicate detection and matching); \texttt{f\_analysis\_assort} and \texttt{f\_analysis\_price} (sample membership flags); \texttt{region\_final} and \texttt{region\_source} (assigned region and its provenance: municipality lookup, platform label, FSA-centroid point-in-polygon, or unknown); \texttt{category\_model} (classifier label, Amendment A-002); \texttt{comparable} (variant-structure comparability of a matched pair).78\section{Category taxonomy and audit details}\label{app:categories}9\input{tables/category_audit}1011\section{Within-region concentration}\label{app:concentration}12\input{tables/h3_within_region}1314\section{Region-level indicators}\label{app:indicators}15\input{tables/h7_indicators}16\input{tables/qdaci}1718\section{Preregistration, decision log and amendments}\label{app:prereg}19The frozen preregistration (\texttt{research/preregistration.md}), the decision log (\texttt{research/decisions.md}, D-001 to D-007) and the two amendments are reproduced in the repository. A-001 (region assignment) replaced ``platform label first'' by ``official geography first'' after a first attempt with the Statistics Canada 2021 economic-region layer produced spurious Estrie/Montérégie conflicts that predate the 2021 boundary change; the final rule uses the MRNF municipality table for urban forward sortation areas, then the platform label, then the FSA-centroid point-in-polygon. A-002 (category labels) introduced a supervised classifier after the audit measured the keyword rules at 46\% precision, with the primacy rule stated before any result was computed under the new labels. Three adversarial referee reports (econometrics, economic geography, data engineering) and the point-by-point response log (\texttt{reports/referee\_responses.md}) are part of the replication package; the corrections they triggered --- a re-projected wild-cluster-bootstrap disturbance, a truncated log-normal fit, null-calibrated cell counts, consensus-based matching precision, the complete listing history for the dynamic exercise, the 2021 boundary transfer applied to platform labels, and a province-wide-sources variant --- changed the status of H5-T1 and H1 and are reported as such (decision D-009).20\section{Prospective collection plan}\label{app:plan}21Three changes to the platform would turn the cross-section into a research panel without changing its public behaviour: (i) a \texttt{price\_log} table recording every observed price per listing and sync, with the regular price and the variant identifier, so that repricing, promotions and reference prices become observable; (ii) product-level provenance fields populated from the certification marks that already exist at the enterprise level (``Produit du Québec'', ``Fabriqué au Québec'', ``Conçu au Québec'', Aliments du Québec), attached to listings only when the merchant's catalogue carries the mark, which would make H6 testable on a subset; (iii) a civic-address geocoding step with retained precision flags, replacing FSA centroids. A repeated snapshot every quarter under the same cleaning rules would allow the related-diversification question (H4) to be asked at the region level, and a second aggregator or a probability sample of merchants from the ISQ business register would allow the coverage ratios of Section~\ref{sec:data_repr} to be turned into selection weights.22\end{appendices}23