The Resolution Layer · Case Study in BriefEKOM
Sewing & embroidery machines.
A cold read of the machine listings of an online sewing-machine retailer — where a maker's machines come in families a few features apart, and a buyer chooses between siblings by the numbers. One of a set of anonymized EKOM analyses — what surfaced, and where it leads.
Lead finding · the page describes another machine
Thirteen listings carry copy written for a different model.
A machine near the top of its maker's range is described, from its first sentence, as a sibling model. Another opens by selling a different machine by name. Headings name a neighbouring model number under titles that name the right one, on new and open-box listings alike. The storefront publishes each description in its structured product data, so the assistant or search engine reading the record learns the sibling's name along with the price.
Also surfaced
Specifications that disagree with themselves
Stitch memory in one field a tenth of the figure in the other two. Presser-foot pressure less than half. A quarter-inch hemmer that makes a half-inch hem. Thirteen findings where the same fact is written twice and the copies do not agree.
Two conditions for one machine
Thirty-nine listings titled open box that explain on the same page they are factory serviced; refurbished listings tagged factory serviced; and no condition field anywhere for a feed to read.
Weights that are defaults
The two most common shipping weights are zero and exactly one thousand pounds. Seventy-four listings that ship weigh nothing; 208 weigh a thousand pounds, down to a steam press whose own page gives under a tenth of that.
The finding that did not survive
Eight flagged "wrong photograph" findings were withdrawn after the photographs were opened: they are right. What is wrong is the file names — the lead image on more than four hundred listings is filed under another brand's product.
How EKOM read it. The engine read the retailer's own public product feed cold — no file, no credentials, no schema — and re-checked every headline finding against the live store more than two weeks later. It profiled what each field is for from how the catalog itself uses it, then held every listing against itself and its siblings: the heading against the title, the description against the specifications, the condition label against the body copy. Where a finding failed that check, it was withdrawn and says so.
From diagnosis to channel-ready. Most of the 74 findings collapse into six upstream mechanisms — new listings built from a sibling's page, one fact kept in several fields with nothing holding them together, condition written into the title, weight left at a default, no enforced list for type and brand, and a bulk import that re-keyed records. Fixing the findings fixes the findings; fixing the mechanisms fixes every listing that has already passed through them and every one that will. Then the same intelligence holds the line at intake, so the next page built from its sibling's is caught the day it goes live.