Fashion: the variant explosion - sizes × colors × fits eating your catalog alive
Every fashion catalog eventually does the multiplication: 8 sizes × 12 colorways × 3 fits = 288 sellable items - of one style. Multiply by a 200-style catalog and you're operating tens of thousands of variant-level records, against Shopify's 100-variant ceiling, feed specs that want per-variant rows, and customers who just want the navy one in medium. This is fashion's defining data headache, and the playbook for it pulls together threads from across this academy into one industry-specific answer.
The modeling answer: split by color, size within
The Phase 1 decision, resolved for fashion specifically: one product per colorway, sizes (and usually fits) as variants. Why this split wins: it respects the ceiling (8 sizes × 3 fits = 24 variants, room to spare), it matches shopping behavior (customers browse colorways as distinct items - each deserves its own URL, images, and ranking life, per the canonicalization judgment call from Phase 5), it produces clean feed structure (colorway products grouped by item_group_id, Phase 7's variant lesson working as designed), and it gives merchandising real units to work with. The connective tissue that makes the split operable rather than fragmenting: a style-level identifier (a "style code" metafield shared across colorway products) so that style-level operations - enrichment, seasonal moves, reporting - can address "this style, all colorways" as one thing. Without it, the split becomes the fork-multiplication problem from Phase 9; with it, you get both granularity and coherence.
The data answer: inherit down, override up
288 items do not need 288 hand-written data sets. The structure: style-level facts (material, construction, fit description, care, sustainability claims) authored once and inherited by every colorway and variant; colorway-level facts (color family, images, color-specific notes) per colorway product; variant-level facts (size, per-size measurements, GTIN, stock) at the bottom. Enrichment effort then scales with styles, not variants - the only arithmetic that survives a large catalog (Phase 10's enrichment loop, running at the right level). The QA rules from Phase 10 get fashion-specific teeth here: every colorway of a style carries identical style-level values (drift between colorways of one style is the classic consistency failure), every variant has its size in the structured option (never the title), every colorway has its own image set.
The survival answer: watch the seams
Where variant explosion actually burns stores, and the standing checks for each: the ceiling (styles approaching 100 variants need the split before the season that breaches it - an audit query, quarterly); duplicate-content gravity (twelve colorway descriptions that are 95% identical - style-level copy inherited verbatim needs colorway-level differentiation on at least title, color language, and imagery, or Phase 4's supplier-description problem recurs internally); feed size and freshness (tens of thousands of variant rows stress sync cadences - Phase 7's staleness math, rechecked at your real volume); and filter integrity (color and size filters run entirely on this structure being clean - one "Navy blue" among the "Navy"s and the facet splits). None of this is exotic; all of it is the academy's standard machinery pointed at fashion's specific multiplication. That's the pattern for this whole phase: every industry headache is general principles meeting a vertical's particular arithmetic.