1\section{Introduction}\label{sec:intro}23Regional economies are usually measured through what is produced, employed or exported. The rise of small-merchant e-commerce has created a different and almost unexploited record: the product catalogues that independent businesses publish on their own storefronts. For a province such as Quebec --- with a distinct consumer-nationalism movement (``acheter québécois''), provincial certification marks, a dense network of agri-food and craft associations, and seventeen administrative regions that differ enormously in size and remoteness --- these catalogues raise a natural question: can product-level online catalogues reveal the spatial organisation, diversity, specialization and relatedness structure of a regional retail economy beyond what business counts show? This paper answers that question for one snapshot (1 November 2026) of Fabri-Ka, an aggregator that indexes the public catalogues of 1{,}569 Quebec-based online stores, and it does so with a deliberately sceptical design: every regional statistic is confronted with a null that reallocates merchants rather than listings, every category label is audited and re-estimated, every matched product is validated, and the hypotheses, estimands and decision rules were preregistered before computation.45The answer is nuanced, and two of the paper's first-draft conclusions did not survive the adversarial refereeing of its code, which we report rather than hide. Online assortments do carry a regional signal, but it lives in the population of merchants, not in the volume of listings. Under merchant-equal weights --- each store counts once --- the dependence between region and product category is far stronger than in 2{,}000 random reallocations of stores across regions (Cramér's $V$ of 0.112 against a null mean of 0.086, $p<0.001$), it holds at the level of regions and of MRCs, under both category labellings, and when every store found through a region-specific directory is dropped; the pattern is readable under the primary labelling: jewellery, fashion and children's goods in Montréal, alcohol and maple in Estrie, pets and alcohol in Montérégie, fashion in Gaspésie, outdoor goods and art in Capitale-Nationale, chocolate in Outaouais. Under listing weights, however, the same statistic is indistinguishable from the null ($p=0.20$, and $p=0.43$ after trimming the largest one per cent of stores): a region's catalogue, counted in listings, is mostly the catalogue of whichever large merchants are located there --- although this is a low-power comparison, because four stores hold a quarter of the regional sample's listings and the null is correspondingly wide. By the joint criterion we preregistered, H1 is not supported; the merchant-level dependence is a descriptive finding. Catalogue sizes are extremely skewed --- a Gini of 0.85 and the top decile of stores holding 79\% of listings --- which is what makes the two weightings diverge. Total regional variety rises with regional size, but the breadth of a typical set of fifteen merchants (0.73--0.89 of its maximum) shows no detectable association with population, business counts, income or density across sixteen regions, a null that the sample size can only bound loosely. On matched products sold by several merchants (1{,}064 groups, matching precision 0.98--1.00), posted prices differ by a median of 4.7 log points and by more than five per cent in half of the groups; resellers and collective storefronts post prices 11.6 log points below the producers' own shops for the same item, a result that survives every correction, whereas the joint test of regional price levels --- significant in a first draft --- does not reject equality once the wild cluster bootstrap re-projects its disturbance off the product fixed effects, and rests on a single region with one large reseller. An economic-complexity indicator built from the region--category matrix fails its preregistered validation: its ranking is unstable under resampling of merchants (median Spearman correlation 0.68 with the full-sample ranking, against a criterion of 0.8), and we therefore withhold both the complexity component and any composite score, publishing the Quebec Digital Assortment and Complexity Index as a family of separately-validated components instead.67Measurement is half of the paper, by design. The platform's keyword-based category labels agree with annotator consensus on only 46\% of a stratified audit sample; a supervised classifier trained on model annotations and validated on the held-out audit reaches 88\% (91\% listing-weighted), and by a rule fixed before any result was computed under it, becomes the primary labelling, with the preregistered rule labelling retained in every table. Merchant geography had to be rebuilt from official boundaries: a first attempt with Statistics Canada's 2021 economic regions produced spurious conflicts that predate the 2021 transfer of two census divisions to Estrie, the platform's own labels carry that pre-2021 map for sixteen stores, and 19\% of listings belong to merchants whose location could not be established. A representativeness audit against ISQ business counts shows that the aggregator's coverage tracks the regional business base closely in levels but varies six-fold per thousand businesses across regions, because the directories that feed discovery are themselves regional. These are not footnotes: they bound every interpretation, and the paper's main methodological contribution is to show how far a non-probability catalogue source can be pushed when each of these defects is measured rather than assumed away.89The paper relates to, but is distinct from, four literatures (Section~\ref{sec:lit}): the measurement of regional specialization and localisation \citep{ellison1997,duranton2005}, whose dartboard logic we transpose to merchants; economic complexity and relatedness \citep{hidalgo2007,hidalgo2009,mealy2019}, whose tools we subject to validation on an unusually small and unusually based matrix; the economics of online retail, the long tail and the geography of online demand \citep{brynjolfsson2011,quan2018,forman2009}, which we mirror from the supply side; and the use of scraped prices and non-probability data \citep{cavallo2017,meng2018}. No method here is new; the object, the validation discipline and the honest accounting of what the source cannot say are. Section~\ref{sec:setting} describes the institutional setting and the platform, Section~\ref{sec:data} the data construction and audits, Section~\ref{sec:hyp} the preregistered hypotheses, Section~\ref{sec:methods} the methods, Section~\ref{sec:results} the results, Section~\ref{sec:robust} the robustness and external validation, and Section~\ref{sec:discussion} the interpretation and limitations; Section~\ref{sec:conclusion} concludes with a verdict on each hypothesis.10