1\section{Conceptual framework and preregistered hypotheses}\label{sec:hyp}23\paragraph{What a catalogue can and cannot reveal.} Let $s$ index merchants, $r(s)$ the administrative region of the merchant's business location, and $c$ one of twenty product categories. A listing is a product a merchant chooses to publish online at the snapshot date. A region's online assortment is therefore the union of the publication decisions of the merchants located there, filtered by the aggregator's discovery process. Three forces shape it: the regional production and retail base (which merchants exist), the decision to sell online and to publish a catalogue (adoption), and the aggregator's coverage (discovery). Only the first is of economic interest; the other two are nuisance and are the object of the representativeness audit. Because the unit is a listing and not a sale, an employee or a shipment, all quantities below are \emph{assortment} quantities, and because merchant location is not product origin, all regional statements concern merchants based in a region, not goods made there.45\paragraph{Hypotheses.} The hypotheses, estimands, specifications, exclusions and decision rules were frozen in a preregistration document before any Phase~4 statistic was computed (SHA-256 recorded in the decision log; two amendments, A-001 on the region-assignment rule and A-002 on category labels, are reported in Section~\ref{sec:data} and were decided before seeing results under the amended definitions).67\begin{description}[leftmargin=1.2em,style=nextline]8\item[H1 --- Regional specialization (confirmatory).] Assortment composition differs across regions beyond what a random allocation of merchants to regions would produce. Statistic: Cramér's $V$ of the region$\times$category table; null: 2{,}000 permutations of region labels across stores. Decision: $p<0.05$ (Holm-adjusted) under both merchant-equal and trimmed-listing weights.9\item[H2 --- Diversity and scale (confirmatory, low power).] Rarefied assortment diversity (entropy at fifteen stores) is associated with regional population. Statistic: Spearman $\rho$ with a bootstrap interval; supported if the interval excludes zero.10\item[H3 --- Concentration (confirmatory).] Catalogue sizes are highly skewed: Gini $\ge0.70$ and top-decile share $\ge0.50$, with bootstrap intervals.11\item[H4 --- Related diversification (exploratory).] Within-store entry into categories absent at baseline increases with relatedness density; region-level entries are counted and declared untestable if fewer than ten.12\item[H5 --- Price dispersion (confirmatory, two tests).] For matched products, posted prices are (T1) geographically structured --- joint wild-cluster-bootstrap test of region effects --- and (T2) competitively structured --- reseller versus producer effect.13\item[H6 --- Local-origin premium.] Declared \emph{untestable} ex ante: the source has no product-level provenance. A store-certification contrast is reported descriptively and labelled as not being H6.14\item[H7 --- Digital-assortment complexity (validation).] A region--category complexity indicator is reported only if it passes bootstrap rank stability ($\rho\ge0.8$), a store-reshuffle null, and a leave-one-region-out predictive check; otherwise it is reported as unstable and excluded from the index family.15\end{description}1617The confirmatory family is $\{$H1, H2, H3, H5-T1, H5-T2$\}$ with Holm step-down at $\alpha=0.05$; raw $p$-values are also reported. Every causal reading is excluded by design: the data are a single cross-section of posted listings from a non-probability discovery process.18