The Resolution Layer · Case Study in BriefEKOM
Promotional drinkware.
A cold read of the drinkware category of a branded-merchandise supplier that sells through distributors — where a distributor quotes from the page, and a search engine or an assistant reads a second copy written for machines. One of a set of anonymized EKOM analyses — what surfaced, and where it leads.
Lead finding · the copy machines read
On all but one product, the machine description has its punctuation stripped — "18/8 stainless" reads as "188".
Each product page carries a record for search engines and assistants, and its description is the page's text with the hyphens, slashes, percent signs and apostrophes removed: "double-wall" becomes doublewall, "100% recycled" becomes 100 recycled, and a stainless grade becomes 188. Hyphens drop out on most products. The same template leaves the brand field blank on all 233 products carrying the supplier's own house brands, and writes the price in a form a standard parser may not read. Nothing in that record is rare or hidden — it is the copy a search engine or an assistant reads first.
Also surfaced
A 32 oz mug, specified as 20
Four products carry a capacity field that contradicts their own name, one by 12 ounces — and 52 more state a capacity in the name and carry none, so a capacity filter cannot find them.
A tumbler 433 inches wide
Three physically impossible dimensions — a 433-inch width and two shipping cases of 148 and 118 inches — and five records with a dimension missing. The only three impossible values in the category.
Lids that are included, optional and "No"
Ten products list an included and an optional lid; twenty list an included lid with no value behind it; seven gift sets and bottles read "No" beside a twist-on closure.
Insulation typed three ways
Hours stated in the copy but missing from the fields, and one insulation type written three different ways — so a filter on one spelling misses the rest.
How EKOM read it. The engine read the supplier's own public product pages cold — no file, no credentials, no integration — and re-read every product named in the findings live the same day. It held each product against itself: the name against the capacity, the features against the lid field, the copy against the hours, the page's description against the structured data written for machines. Every extent is an analyst's recount of the archived pages, not the automated pass's own count.
From diagnosis to channel-ready. The 23 findings read as four upstream mechanisms — a separate template that writes the machine copy, two sources that disagree with nothing to referee them, a default standing in where a value should be, and typed values with no plausibility check. Fixing the findings fixes the findings; fixing the mechanisms fixes every product that has already passed through them and every one that will. Then the same intelligence holds the line at entry, so the next 433-inch width is held for review the day it is typed.