1\section{Related literature and positioning}\label{sec:lit}23The paper draws on six literatures. We summarise what each establishes, and where the present exercise sits relative to its closest empirical precedents. Every reference below was verified against Crossref or the publisher's page before citation; one critique of the method of reflections that could not be verified is not cited.45\paragraph{Economic geography and regional specialization.} Industries are localised far beyond a random allocation of plants across space \citep{ellison1997,duranton2005,mori2005}, with heterogeneity across industries and scales \citep{rosenthal2001} that also holds in Canadian plant microdata \citep{baldwin2008}. The micro-foundations are sharing, matching and learning \citep{marshall1920,duranton2004}; density elasticities of productivity are modest and demanding to identify \citep{rosenthal2004,combes2015,duranton2020}. Whether local specialisation or diversity favours growth remains contested \citep{glaeser1992,henderson1995}, and the related/unrelated-variety decomposition of entropy is the modern operationalisation \citep{frenken2007,content2016}. Market size and love of variety \citep{krugman1980,krugman1991,fujita1999} and the skill-ordered comparative advantage of cities \citep{davis2020} motivate a link between regional scale and assortment breadth; urban scaling shows that the number of distinct business types grows roughly with the logarithm of population and that activities appear in a nested order across city sizes \citep{bettencourt2007,youn2016,hong2020}. For Quebec and Canada, \citet{polese2004,polese2006} and \citet{shearmur2007,shearmur2008} document a persistent centre--periphery gradient in employment location. None of this work uses the catalogues of online merchants as the object of measurement, and none covers Quebec's seventeen administrative regions in the 2020s. Our revealed assortment advantage (RAA) is the Balassa logic \citep{balassa1965} applied to listing counts, with the dartboard idea of \citet{ellison1997} supplying the null: a random allocation of \emph{merchants}, not listings, to regions.67\paragraph{Economic complexity, relatedness and diversification.} The Balassa index is skewed and unstable at the top, and symmetric or normalised variants exist \citep{hinloopen2001,yu2009,laursen2015}. The product space \citep{hidalgo2007} and the principle of relatedness \citep{hidalgo2018,boschma2017} are robust regularities: regions diversify into activities related to what they already do, documented with industries \citep{neffke2011}, patents \citep{kogler2013,rigby2015,balland2017}, occupations \citep{muneepeerakul2013} and firms \citep{gao2018}, and combined with complexity in the smart-specialisation framework \citep{balland2019}. Economic complexity and fitness indices summarise bipartite region--activity matrices \citep{hidalgo2009,hausmann2011,hausmann2014,tacchella2012,cristelli2013,servedio2018}; the ECI is formally a spectral clustering of a region-similarity matrix \citep{mealy2019}, rankings are sensitive to algorithm and sample \citep{morrison2017,sciarra2020}, and structure in bipartite matrices must be tested against degree-preserving null models \citep{saracco2015}; see \citet{hidalgo2021} and \citet{balland2022} for surveys. All applications we could verify use production-side bases (exports, patents, industries, occupations, firms). Nothing is known about the behaviour of these tools on a $17\times20$ matrix built from online listings, which is why we treat the complexity indicator as an exercise in validation rather than as a ranking (Section~\ref{sec:results_h7}).89\paragraph{Online retail, the long tail and the geography of online demand.} Online variety created consumer surplus and shifted sales toward niche products \citep{brynjolfsson2003,brynjolfsson2011,zentner2013,waldfogel2017,anderson2006}, but hits remain dominant and the tail is thin in revenue \citep{elberse2008,tan2017}; search-cost theory predicts both long tails and superstars \citep{barisaac2012}. E-commerce reallocates retail share toward large firms and platform owners \citep{goldmanis2010,zhu2018,einav2016}. The geography of online demand is systematic: consumers buy more online where local supply is thin and when they are preference minorities \citep{forman2009,choi2011,brynjolfsson2009}, demand heterogeneity across markets limits the value of online variety for products with local tastes \citep{quan2018}, distance still matters online \citep{hortacsu2009,blum2006,lendle2016}, gains are larger for remote consumers \citep{dolfen2023,couture2021}, and e-commerce changes relative prices across space \citep{jo2019,lieber2012,goldfarb2019}. Platform design shapes dispersion \citep{dinerstein2018,ellison2009,baye2004,baye2006,brynjolfsson2000,clay2001}. This stream is demand- or transaction-based. The supply-side question --- what independent local merchants put online and how it varies across the regions of one province --- is, to our knowledge, not addressed; \citet{quan2018} and \citet{forman2009} are the closest mirrors, and our data cannot say whether regional assortments reflect regional tastes.1011\paragraph{Regional inequality and digital adoption.} Firms in large cities adopted the commercial internet first \citep{forman2005}; the wage gains from internet investment accrued to already-advantaged places \citep{forman2012,sinai2004,forman2018}; broadband effects on rural growth are positive but modest and hard to identify \citep{kolko2012,whitacre2014,haller2015,haller2019}; adoption and usage divides differ \citep{goldfarb2008}; Canadian household divides are documented with the Canadian Internet Use Survey \citep{haight2014,statcan_cius,statcan2023cius} and business digital adoption with the SDTIU \citep{statcan_sdtiu}; the OECD measurement framework is \citet{oecd2019} and the Quebec programme is \citet{atn_netendances}. We found no peer-reviewed study of regional e-commerce \emph{supply} in Canada or Quebec; the Quebec-specific evidence is grey literature \citep{isq2024}, and the paper is explicit that this stream is the thinnest in verified academic sources.1213\paragraph{Product matching, hedonic prices and spatial price dispersion.} Record linkage rests on \citet{fellegi1969} and \citet{christen2012}. Scraped online prices are a valid but selected source: online and offline prices coincide for most multi-channel retailers \citep{cavallo2017}, scraped data have specific frequency and stockout properties \citep{cavallo2018,cavallo2023}, and the Billion Prices Project set the standards \citep{cavallo2016}; online price-setting and cross-border integration are documented by \citet{gorodnichenko2017,gorodnichenko2018,cavallo2014}; web-scraped prices now enter official indices \citep{konny2022}. Dispersion is mostly relative and persistent \citep{kaplan2015,kaplan2019}; US chains price almost uniformly across locations \citep{dellavigna2019}; spatial price gaps grow with distance \citep{atkin2015}; variety and prices differ across cities \citep{handbury2015,handbury2021}; hedonic theory and practice follow \citet{rosen1974,pakes2003,bajari2005}. Cross-merchant dispersion among many \emph{independent} small online stores in one province, with product fixed effects from matched titles and SKUs, has no direct precedent; the near-collinearity of merchant and region effects is specific to our setting.1415\paragraph{Measurement error and representativeness of non-probability data.} For non-probability big data the error is governed by the data-defect correlation rather than by sample size \citep{meng2018,bradley2021,lazer2014}; inference requires explicit selection models \citep{elliott2017,buelens2018,heckman1979,groves2006,little2019}, and statistical agencies are explicit about the conditions \citep{daas2015,beaumont2020,rao2021}; measurement error in regressors and outcomes \citep{hausman2001,bound2001} and the modifiable areal unit problem \citep{briant2010} are additional threats. We take these literatures literally: Section~\ref{sec:data} reports a data-quality ledger, a category-label audit, a matching validation and a representativeness audit against Statistics Canada and ISQ counts.1617\paragraph{What is and is not new here.} The contribution is a documented, validated supply-side dataset of online product listings of Quebec-based merchants by administrative region; RAA and diversity measures with merchant-count-preserving permutation nulls and merchant-weighted variants; a region--category bipartite indicator whose eigenvector interpretation is tested against bootstraps and null models and reported as a family of components rather than a headline score; cross-merchant dispersion of matched products among independent stores with explicit collinearity diagnostics; and a representativeness audit of an aggregator against business counts. No method is new. We do not provide evidence on agglomeration economies, on diversification over time, on sales concentration, on gravity in online trade or on welfare, and we never call the indicator a production or economic complexity index: without verified product origin it describes observed online catalogues only.18