The Resolution Layer · Case Study in BriefEKOM
A fashion footwear and apparel brand.
A full-catalog, no-schema read across women's and men's footwear, a clothing line, and accessories — one of a set of anonymized EKOM analyses. What surfaced, and where it leads.
Lead finding · discovery & trust
Product pages serving a photograph of a different shoe.
The brand names its image assets to a strict internal convention — style, then colorway, then sequence — and that consistency is what makes the exceptions readable. Six separate findings, all critical, where the asset on the page is named for a different style entirely: every colorway of one dress-shoe style served another style's photograph; a second style served a third style's; a skirt served a different colorway's. On one sneaker family the convention is abandoned altogether for generic upload names, where every sibling colorway conforms.
Also surfaced
A fiber content that cannot be true
Both colorways of one dress state their composition as 55% rayon 45% rayon — the same fiber named twice at two percentages. Fiber content is federally required to be accurate on US apparel.
Leather in the name, suede in the spec
Two leather colorways open their specs with Suede upper material while their own sibling colorway correctly says Leather upper material. The name promises one material; the spec on the same page states another.
Color values cut off mid-word
Two styles carry BLACK LEAT instead of BLACK LEATHER, dropping out of color filtering; another carries a duplicated, part-truncated value that renders a phantom second swatch leading nowhere.
One unstructured field, eight defects
Heel height, materials, lining, sole and country of origin all live in one free-text string — so heel eight, heel heigh, a boot measured around twice and along never, and a bag with two heights all survived into the live catalog.
How EKOM read it. With no schema and no manual setup, the engine profiled the catalog, learned the brand's own image-naming convention, and read the catalog in full rather than sampling — which is what made a filename naming the wrong style legible as a defect instead of noise.
Shown, and shown honestly. Of 40 flagged items, six came from a heuristic layer. Four were withdrawn on review — two describe how EKOM read the catalog rather than anything wrong inside it, one is not a defect at all, and one was a second sighting of a finding already reported. The other two are real, and are kept as structural recommendations rather than counted as defects. The actionable count is 34.
From diagnosis to channel-ready. This pass read and diagnosed. The same understanding powers what follows — correcting the values that have one right answer, structuring the spec field that lets the whole class of defect happen, and holding the line at intake as new styles and colorways arrive.