Resolution Layer Case Study · Read & Diagnose
Heating, cooling & water-heating equipment.
A catalog read across the public product pages of an equipment manufacturer that sells through distributors and contractors — where a model number says one thing, and the specification beside it says another.
This is a real EKOM catalog analysis, with the client's identity removed. The client is a manufacturer of heating, cooling and water-heating equipment that sells through distributors and independent contractors rather than to the public. EKOM read the product pages it publishes in its own sitemap, flagged 64 items, and re-read most of the same pages live two weeks later. 44 are reported here as actionable catalog defects. The rest were folded into findings already counted, held because the live pages no longer showed them, or held at low confidence. None is in any count below.
What makes a manufacturer's catalog different is where it sits. A retailer's product page is one description of a unit among many. A manufacturer's is the one the others are checked against — by the distributor building a listing, the contractor quoting a job, and, increasingly, the assistant asked which heater runs on propane. And nobody chooses this equipment by looking at it. They choose it by its fuel, its capacity, its dimensions and its model number.
In most categories a wrong field costs a click. Here it can put a propane unit on a natural-gas job — and the error starts at the source every downstream catalog copies.
Product-line names, model numbers and field values that could identify the manufacturer have been removed or generalized. Counts of affected models are exact.
What's inside
- At a glance — the severity split, where the findings concentrate, and how the read was done.
- What an installer runs into first — propane units filed under natural gas, and page headings that name a different model.
- The receipts — specifications that cannot be right for the unit, and placeholders printed as values.
- What a machine is told — the layer search engines and assistants read before the page.
- Where it traces back, and what's next — the five mechanisms most of these findings collapse into, and the order to fix in.
Nothing was supplied and no internal system was touched. Everything below was read from the manufacturer's public product pages.
44
Actionable findings
(64 flagged)
Where the 44 concentrate
Theme
What it is
Findings
Heading vs. model
A page heading naming a neighbouring model family
7
Impossible specifications
Heights, tanks and tonnage from another size
7
Placeholders as values
Stand-in numbers, "NULL", an editor's note
6
Missing specifications
No efficiency rating, voltage or tonnage
6
Fuel type
Propane units filed under natural gas
5
Machine layer
A flat breadcrumb; dead and doubled sitemap entries
4
Values, marks and identifiers
Split colour values; crossed trademarks; shared SKUs
9
How EKOM read this catalog
1
Read
Every product page in the manufacturer's own sitemap, fetched as a search engine or assistant fetches it — no file handoff, no credentials, no integration.
Each page is one model: one SKU, one specification block.
→
2
Profile
Infer what each field is for from how the catalog itself uses it, then hold every model to that.
The model number is the most reliable field on the page; most findings test the rest against it.
→
3
Analyze
Surface defects, rate severity, and bind each one to the models and values behind it.
44 findings, 9 themes, 5 causes.
Machine-surfaced signals from a cold read, checked by an analyst and re-read live — not a verified defect list. Some will prove intentional, which is why each is tied to the values behind it.
What an installer runs into first.
The lead finding is about fuel type — on a gas appliance, a safety and compatibility attribute, and the first filter anyone applies to a water heater.
Critical · Fuel type contradicts the model
The model number says propane. The fuel field says natural gas.
On a line of commercial gas water heaters, the propane models carry the propane designation in both the model number and the page address. Their fuel field disagrees. A contractor filtering for propane does not find the ones that say natural gas; one filtering for natural gas does.
Models re-read live
What the fuel field holds
Models
Propane
natural gas only
6
Propane
both natural gas and propane
39
Propane
propane alone
0
Natural gas
natural gas only
28
Not one propane model lists propane alone, and not one natural-gas model lists propane at all. The propane value exists elsewhere in the catalog, so the vocabulary is there. If these units ran on either gas, their natural-gas twins would say so too; they do not. The record cannot be right both ways.
Fuel type decides the gas valve, the orifice and the supply line. A propane unit presented as natural gas is the kind of error found at start-up rather than at the order. Two more findings sit in the same theme: gas furnaces and packaged gas-electric units that publish no fuel at all.
A heading that names another model.
The page title and the SKU name the model. On a run of model families, the page heading — the largest text on the page, and the one a parser takes as the product's name — names a different one.
Critical · Heading, title and SKU disagree
An air handler whose heading names its neighbouring family.
On one air-handler model the SKU and the page title agree. The heading carries the name of a different model family, and the page address agrees with the heading. A contractor quoting one model lands on a page headed with another, and anything that takes the heading as the product name records the wrong model — with the address seeming to confirm it.
Critical · The same shape, catalog-wide
213 pages in seven model families, each headed with a neighbour's name.
Compared field by field on the live re-read, the heading is always the name of a neighbouring family, never a random string, while title and SKU name the model itself. That pattern reads like a template taking its heading from the family page a model is filed under. The same shape fits a line of point-of-use water heaters whose title and heading state opposite versions of the one feature a buyer of that product is choosing on, and a furnace filed under a sibling series' page title.
Seven separate findings in the analysis turned out to be this one behaviour. That is the difference between correcting 213 headings and changing where a model page takes its heading from.
Numbers a specifier would act on.
These are values in the specification block itself — the figures a contractor reads to size a system, fit a curb or match a replacement — and each one cannot be right for the model it sits on.
Specification
What the page says
Severity
Height, a commercial packaged unit
Five inches, where its sibling sizes run to several feet — a dropped digit is the likely reading
critical
Tank, a heat pump water heater
A 65-gallon model — 65 in its own model number — publishing the 50-gallon tank of its smaller sibling
critical
Tonnage and capacity, a packaged unit
The tonnage and capacity of the next size down, while every sibling of its own size agrees on the right figures
critical
Maximum tonnage, a packaged unit
A maximum equal to its own minimum, under a title stating a wider range
critical
Efficiency, an evaporator coil
A seasonal efficiency rating of its own — a rating that belongs to a complete system, not a coil
critical
Height, two water heaters
One inch
high
High · A unit's range is the series' range
A 3½-ton condenser that publishes a capacity range across the whole series.
On the large majority of pages that carry a unit's own tonnage alongside a minimum and a maximum, the minimum and maximum read as the range of the series, not of the unit. As published, a filter for units that reach the top of the series returns a smaller model, and so does one for units that go as low as the bottom of it.
Most of these were found by comparing a model with the sibling that gets it right, and the right figure is often written in the model number itself. That makes them corrections to confirm, not research to commission.
Filled in is not the same as right.
One idea about the findings rather than more findings — because it explains why most of this would never appear on a fill-rate report.
Most of what this study found is not blank. The fuel on the propane heater is filled in. The heading on the air handler is filled in. The tank on the 65-gallon unit is filled in. A completeness check — the instrument most catalogs are measured with — scores every one of those fields at a hundred percent, and every one of them gives the wrong answer.
Critical · Placeholders published as values
A tankless heater credited with a hot-water figure no heater has.
Ten reduced-voltage variants of a tankless electric line publish a database stand-in number in two hot-water delivery fields — fields that do not apply to a tankless unit at all. On the same pages, a display label has been written into the SKU itself, so a distributor matching on the model number may not link it; and the base model's own page carries three specifications where its variant carries nineteen.
The other placeholders
An editor's "(Duplicate)" note in the model name of a packaged heat pump line. The literal text "NULL" as the install location of several gas water heaters. A colour field holding a number, and another holding a misspelling. Each would pass a completeness check. None of them is a specification.
What catches them is not asking whether a field is filled, but checking the value against the model number and against the sibling that gets it right. That is also why the analysis could stay out of the page copy: efficiency figures printed in cross-sell panels describe other products, and a narrow read of the specification block keeps them out of the record entirely.
Beneath the page a person sees sits a layer written only for machines: the structured product data, the breadcrumb trail and the sitemap. Search engines and AI assistants read this layer first.
Observed, not a finding
The manufacturer tells machines what every model is. By design, it does not tell them what any model costs.
Effectively every product page carries structured product data with a name, description, image, brand and SKU — and no offer: no price, no availability. For a manufacturer whose distributors and dealers set the price, that is the expected architecture, and the study does not count it. It does decide what a machine can learn here: identity from the manufacturer, everything commercial from someone downstream. That makes the identity fields the part of the page that has to be exact.
Critical · Machine layer out of step
People see a five-step category path. Machines are told two steps, ending in "product".
The visible breadcrumb on a product page runs from the home page down through the product type to the model family. The machine-readable breadcrumb on the same page is a generic two-step value, and the analysis found it throughout the catalog. The category structure exists — it is on the screen — and none of it reaches the markup a search engine uses to understand where a product belongs.
In the published sitemap
What the re-read found
Count
Dead products
Listed, and returning "not found"
8
Doubled entries
Byte-identical addresses listed twice, all in one coil family
21
A content page
Filed as a product, with no product record and no specifications
1
44 findings sounds like 44 problems. Most of them collapse into five upstream mechanisms — and a mechanism can be fixed once and stopped from recurring. Each tell below is a pattern in the data; the mechanism is EKOM's reading of it.
Model pages that carry the family's text
The tell: the wrong heading is always a neighbouring family's name, and where checked the address agrees with it; the minimum and maximum tonnage read as the series range. The model's own fields — SKU and title — are right. It produced the heading findings and the tonnage ranges.
A fuel value set for the line, not the unit
The tell: every propane model re-read lists natural gas; no natural-gas model lists propane; the propane value is in use elsewhere. It produced the propane findings.
Values taken from the neighbouring size
The tell: the wrong figure is a real figure belonging to a sibling — the smaller tank, the next size down's tonnage, a maximum equal to its own minimum.
Database and editor defaults exported as content
The tell: the same value regardless of what it describes — a stand-in number, "NULL", "(Duplicate)", a colour of "1", a breadcrumb and a share title that both read "product".
A publishing layer generated apart from the product set
The tell: the sitemap lists pages that no longer exist, lists some twice and lists a content page as a product; voltage variants are published as their own pages with the label inside the SKU.
Correcting 213 headings by hand corrects 213 pages. Changing where a model page takes its heading from corrects those and every model filed under a family page after them.
33%
In a survey of 750 professional buyers across manufacturers, wholesalers, distributors and retailers, a third reported order errors caused by inaccuracies on a supplier's web store, and three in four said they would switch suppliers for a smoother online experience. In this catalog the inaccuracies sit exactly where a model is chosen: the fuel, the capacity, the dimensions and the model name.
Sana Commerce B2B Buyer Report, published January 2025; fieldwork by Sapio Research, September 2024
A manufacturer's catalog is the source, not a copy. When a distributor builds a listing, when a contractor's quoting tool looks up a model, when an assistant is asked which heater runs on propane, the manufacturer's page is where a disagreement gets settled. Wherever a downstream catalog takes its data from that page, it takes the error with it — and a copy carries no mark of where it came from.
What this class of defect costs is not a sale — it is the wrong unit. A propane heater filed under natural gas is missing from the propane search and present in the wrong one. An air handler whose heading names another family is recorded as that family by anyone who takes the heading as its name. A 65-gallon unit published at 50 is left out of the search for the size it is. None of these was made at the factory, and each can travel further than the page it is printed on.
A second clock runs underneath. Buyers comparing equipment increasingly start with an assistant rather than a search box, and an assistant reads the fields before anything else: it matches a model, checks the fuel, confirms the capacity. On these pages it finds propane units that say natural gas, headings that name the wrong model and a hot-water figure no heater has. Those are not rankings to trade off. They are wrong answers, and they do not correct themselves.
Three moves, in order. The first needs no new data at all — and one correction that looks mechanical deliberately waits.
This pass read and diagnosed. The same structural understanding powers what follows — turning a diagnosed catalog into one the channel can specify from with confidence.
1 · Apply the corrections the record already answers for itself
Take each page heading from its own model. Clear the stand-in numbers, "NULL" values and editor's notes. Correct the impossible specifications from the sibling or the model number that states the right value. Publish the breadcrumb the page already shows, and reconcile the sitemap. Approved as patterns, not page by page.
2 · Settle the questions that gate the rest — starting with fuel
Setting every propane model's fuel from its model number looks like a mechanical fix, and it waits for one answer: whether any of these units is certified for both gases. So do the family filing, what the minimum and maximum tonnage are meant to describe, and which efficiency metric the rating field holds. Those answers come from engineering first.
3 · Hold the line at the source
Keep ongoing intelligence where engineering data meets the published page, so the next model filed under the wrong family or fuel is caught as it lands rather than found later by a contractor quoting from it.
This is how EKOM moves a catalog from insight to impact —
and keeps every specification reading right as the line grows.