The Resolution Layer · Case Study in BriefEKOM
Safety & personal protective equipment.
A cold read of the safety category of a hardware and home-improvement distributor — where the buyer is not comparing objects they can hold, but a noise reduction rating, a lens tint and whether a mask is a respirator. One of a set of anonymized EKOM analyses — what surfaced, and where it leads.
Lead finding · protection ratings that contradict themselves
One pair of earmuffs states three different noise reduction ratings across four of its own fields.
The product name is built around a 20 dB rating. The description states a third number, 24. One specification key holds a bare 20; its correctly spelled twin holds NRR 27. On hearing protection the number is the product — it is the entire basis on which a buyer chooses — and here the record cannot say which number it stands behind. This is one of only two findings in the analysis that propose no correction at all: there is no majority to fall back on, and the right value has to come from the manufacturer rather than from an algorithm. On its sibling one size along, four fields say 22 and a single key says 23 — and the dissenter is the correctly spelled one, the half of a duplicated key pair that a filter actually reads.
Also surfaced
Four eyewear families holding each other's lens colors
In each pair the two records are identical in every other specification attribute and hold exactly the other's lens color — a name saying Clear against a spec saying Smoke, and the mirror image on the sibling. Random corruption does not produce that shape; an exchange does. Anyone filtering by tint is returned the other product.
Respiratory protection filed under fire extinguishers
An N95 face mask, a disposable paint respirator and a set of replacement pre-filters all carry a byte-identical category path that ends at a fire-extinguisher leaf. Two half-mask respirators sit under a face-shield leaf, and a pair of chemical splash goggles under an earmuff leaf. A shopper browsing respirators never reaches the five respiratory items.
One quantity, two spellings of the same key
A specification key with a letter missing from it sits alongside its correctly spelled twin on seventeen hearing-protection products across the category. Where they agree, the duplicate is merely invisible to a filter. Where they disagree, the misspelled key holds the number that matches the product name — and the key a filter reads holds a higher one.
A fulfillment marker used as a price
The catalog marks drop-ship-only items with a sentinel price, and the same sentinel sits on items flagged as stocked — visible, and with no usable price. The convention repeats in weight, where the placeholder sits on stocked items carrying real prices, so anything quoting shipping from that field quotes against it.
How EKOM read it. The engine took the live public catalog cold — no file, no credentials, no schema — profiled the catalog's own field semantics, then held every record against its siblings and against how each field is used across the category. That context is what turns a valid-looking string into a finding. Here it was the eyewear specification convention: the records share one attribute set and it holds, which is the only reason a swapped value is visible at all. Nothing was told to the engine about what a noise rating should be; it learned which of a product's own fields normally agree, and reported where one stopped.
From diagnosis to channel-ready. The 20 findings collapse into five upstream mechanisms — one column exchanged between two records, two writers for one measurement with one misspelling the key, a fulfillment marker used as a value, a mapping collapsing products onto one wrong leaf, and a vendor's vocabulary passed through unnormalized. Fixing the findings fixes the findings; fixing the mechanisms fixes every product that has already passed through them and every product that will. Then the same intelligence holds the line at intake, so the next contradicted rating never reaches a live page. Every finding here is reported as a factual inconsistency between fields on the same product; no statute is named, and none was named in the analysis.