The Resolution Layer  ·  Case Study in BriefEKOM

Plumbing & heating wholesale.

A cold read of a regional plumbing, heating and HVAC wholesaler's published list-price catalog — a PDF, and the only product record it publishes — where a contractor's quote is built from the list price on the page. One of a set of anonymized EKOM analyses — what surfaced, and where it leads.

Vertical
Plumbing, heating and HVAC distribution — wholesale to the trade
Method
Cold read · the published catalog PDF, nothing supplied · checked against the rendered page
Issues surfaced
8 actionable (44 flagged) · 4 critical
This pass
Read & diagnose
Lead finding  ·  the same catalog number at two list prices
A combination boiler that lists nearly $300 lower in the heating section than under water heaters.
The two sections carry the same manufacturer's table — the same models, the same specifications, word for word — at different prices, and a manifold kit beneath the boilers lists at almost twice the price on one page as on the other. It is not alone: twenty-five catalog numbers are printed at two different list prices, and the disagreements run one way. Wherever a parts page and an equipment page disagree, the parts page is higher; the reprinted heating table runs below the original — the signature of sections repriced at different times and never reconciled. A contractor pricing from one page and a counter clerk reading the other are both right, and they do not agree.

Also surfaced

Prices the tables themselves refute
A mid-size well tank listed at four times the largest tank in its family — the only price in its table without a thousands comma. A blower motor printed as a five-figure number because its decimal point is missing. A one-horsepower motor at ten times the three-quarter-horsepower motor above it. Each was confirmed on the rendered page.
A number that is not an identity
A lavatory drain with three finishes, three prices and one catalog number — the dearest finish at more than twice the cheapest — on a page where fourteen finish-priced lines sit under four numbers. The same catalog puts the finish in the number on its faucet pages. Three other numbers each name two unrelated items.
A format a machine misreads
Prices under a dollar print as the cents alone — .62, .57 — at the end of the leader dots, on more than a hundred lines, most of them in the pipe-and-fittings section. A person reads sixty-two cents; a system reads sixty-two dollars. The analysis's own first pass made exactly that error, which is how the format was found.
Context: a record with one layer
No online catalog, no structured product data, no feed: a PDF typeset for the eye is what every search engine, assistant and quoting tool reads. Not counted as a finding — but thirty-four of the analysis's first-pass signals described reading the document rather than the document, and that is the measure of the format.

How EKOM read it. The engine downloaded the published catalog cold — no file handoff, no credentials — turned its text layer into rows page by page, and profiled what each column is for from how the catalog itself uses it. It then held every line against its siblings: the same catalog number on another page, the next size in a tank ladder, the motor on the line above, the faucet trim whose finish is in its number. Every candidate was checked against the PDF's own text, and the headline values against the rendered page, before anything was written down.

From diagnosis to a record everyone reads the same way. The eight findings collapse into four upstream mechanisms — tables reprinted by hand section by section, values typed rather than generated, an identity convention applied on some pages and not others, and a document typeset for the eye. Fixing the lines fixes this edition; fixing the mechanisms fixes the next one before it is set. Then the same read runs on every new catalog file before it is published, so the next boiler at two prices is caught in the file rather than found by a contractor at the counter.

The Resolution Layer
EKOM
[email protected]  ·  ekom.ai
EKOM
Case study — client anonymized