1\section{Conclusion}\label{sec:conclusion}2Using a documented, read-only snapshot of an aggregator of Quebec online stores, we asked whether product-level catalogues reveal the spatial organisation of a province's online retail economy beyond what business counts show. The verdict on each preregistered hypothesis, after three adversarial referee rounds, is as follows. \textbf{H1} (regional specialization) is \emph{not supported} by its preregistered joint criterion, but the merchant-equal dependence that forms half of that criterion is significant under every sample, labelling, geography and discovery-source variant, at the level of regions and of MRCs: the merchants located in a region publish systematically different categories from a random reallocation of merchants, while listing-weighted composition is indistinguishable from chance --- a low-power comparison dominated by a few very large catalogues. \textbf{H2} (diversity and scale) is \emph{inconclusive}: total variety rises with regional size, the breadth of a typical set of merchants does not detectably. \textbf{H3} (concentration) is \emph{supported}: a Gini of 0.85 and a top decile holding 79\% of listings, with a heavy tail whose family cannot be pinned down. \textbf{H4} (related diversification) is \emph{exploratory}: a positive store-level association between relatedness density and category entry over ten weeks, labelling-sensitive, untestable at the region level. \textbf{H5} is \emph{split}: the channel test is supported --- resellers post lower prices than producers for the same item --- and the geographic test is not, once the bootstrap is implemented correctly and the influence of one region is exposed. \textbf{H6} (local-origin premium) is \emph{untestable} in this source. \textbf{H7} (complexity indicator) is \emph{not validated}: the ECI-type ranking is unstable under resampling of merchants on a $17\times20$ matrix, and the index family is published as components without a composite.34Three lessons travel beyond Quebec. First, catalogue data from independent merchants carry a regional signal, but it is a signal about who sells, not about how much is listed; any regional indicator built on listing counts without a merchant-preserving null will mistake the location of large catalogues for specialization. Second, the measurement defects of such sources --- category labels at 46\% precision, a fifth of listings without a location, six-fold variation in coverage, regional labels that carry a pre-2021 map --- are large but measurable, and measuring them changes conclusions. Third, two of our own first-draft conclusions (a geographically structured price level and a power-law tail) did not survive adversarial refereeing of the code; we report them as withdrawn rather than as findings, which is what a preregistered, auditable pipeline is for. The natural next steps are the ones the source cannot yet support: a price panel with store pinning, product-level provenance from certification marks, civic-address geocoding, and a probability sample of merchants against which the aggregator's coverage could be turned into weights.5