The Resolution Layer · Case Study in BriefEKOM
Eyewear structured data.
A cold read of one premium brand's sunglasses at a specialty eyewear retailer — where every frame has its own page, and the second copy of each product written for search engines is supposed to agree with it. One of a set of anonymized EKOM analyses — what surfaced, and where it leads.
Lead finding · two records, one page
Every page checked embeds two product records for search engines, and they disagree on what identifies the product.
On all five pages opened in a browser, one record comes from the vendor that serves the data and one from the site itself. On a typical page the vendor's record makes the SKU the barcode and the part number the model, and the site's record does the opposite; only the vendor's carries a product group ID. Prices agreed across both records and the visible page. A search engine has to pick one, and the other defects in this read were measured in the vendor's.
Also surfaced
A trademark symbol, written as text
Twenty-five products carry the literal text of the symbol's code point in the data, where the symbol belongs — in the description, and on ten of them in the name and the colour. The visible page shows the symbol correctly; the data behind it does not.
Group IDs cut to six characters
Fifty-one group IDs each span more than one model, covering 690 products — one holds twenty-one products across fourteen models. Forty models are split the other way, across several IDs, and licensed team editions share a prefix across competing teams.
A youth line with almost no data
Twelve products have no barcode field, colour, material, category, gender or sale price, and their part numbers break the catalog's pattern. The barcode is not lost — it is in the SKU and the address — so the gap is recoverable from the record itself.
Findings that did not survive
Twenty were raised and fifteen are reported. Two described EKOM's own conversion of a value, two were not defects, and one was merged into another. None is counted.
How EKOM read it. The engine read every product page of the brand cold — no file, no credentials, no integration — joined the vendor's record to the page, held each product against its siblings, and had an analyst recount every magnitude from the saved records. Five pages were opened in a browser to compare what a shopper sees with what the page embeds. Where a finding failed that check, it was withdrawn and says so.
From diagnosis to channel-ready. Most of the fifteen collapse into six likely mechanisms — two producers of one page's data, an identifier built from a name, a character lost in encoding, product lines set up outside the template, internal codes passed through as display values, and no check at publish. Correcting the findings corrects the findings; correcting the mechanisms corrects every product that has passed through them and every one that will. Then the same intelligence holds the line at publish.