Resolution Layer Case Study · Read & Diagnose
Sewing & embroidery machines.
A catalog read across the machine listings of an online sewing-machine retailer — sewing, quilting, embroidery and industrial machines, where each model is written down in five places and a buyer chooses between near-identical siblings by the numbers.
This is a real EKOM catalog analysis, with the client's identity removed. The client is an online sewing-machine retailer carrying the major household, quilting, embroidery and industrial brands. EKOM read every listing above a price threshold — the machines, and the furniture, presses and software sold beside them — from the retailer's own public product feed, and flagged 95 items. 74 are reported here as actionable catalog defects. Fourteen were withdrawn, eight of them because the automated pass had read image file names as if they were the pictures: the photographs were checked and are right. The rest concerned EKOM's own run, were held back at moderate confidence, were folded into findings that describe the same defect, or were found already corrected. None is in any count below.
What makes this category different is how alike the products are. A maker's machines come in families, a few hundred dollars and a handful of features apart, and a buyer choosing between two of them decides on stitch memory, presser-foot pressure, throat space, and whether the unit is new, open box or factory serviced. Those are exactly the things a product page states in words — and a machine's model is written down in its title, its heading, its description, its specification lines and its SKU.
In most categories a page borrowed from a sibling costs a detail. Here the sibling is the product the buyer is comparing it against.
Manufacturer names and model numbers are not attached to individual defects here, because on a retailer's catalog a maker, a model and a defect together can identify the retailer. The client's prices, SKUs, addresses and field values have been removed or generalized; counts are exact.
What's inside
- At a glance — the severity split, where the findings concentrate, and how the read was done.
- What a buyer runs into first — machine pages that describe a sibling, and specifications that disagree with themselves.
- The receipts — condition labels that contradict each other, and shipping weights that are defaults.
- The finding that did not survive — why eight flagged image defects were withdrawn, and what is wrong instead.
- Where it traces back, and what's next — the six 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 retailer's public product feed, and every headline finding was re-checked against the live store more than two weeks after the first read.
74
Actionable findings
(95 flagged)
Where the 74 concentrate
Theme
What it is
Findings
Copy from another machine
Descriptions and headings that name a sibling model
11
Specifications that disagree
One figure in the description, another in the specifications
13
Shipping data
Zero weights, a thousand-pound default, downloads marked for shipping
12
Category
Machines with no type; industrial filed as household
12
Condition labels
Open box and factory serviced on the same record
5
Identifiers, brand and pricing
SKUs naming another model; "was" prices below the price
8
Page hygiene and the rest
Dead download headings, test tags, file names, duplicate SKUs
13
How EKOM read this catalog
1
Read
The retailer's own public product feed — no file handoff, no credentials, no integration.
Every listing above the threshold, read once and re-checked live later.
→
2
Profile
Infer what each field is for from how the catalog itself uses it, then hold every listing to that.
A model is stated in up to five fields per listing — which is what makes a disagreement between them detectable.
→
3
Analyze
Surface defects, rate severity, and bind each one to the listings and values behind it.
74 findings, 11 themes, 6 causes.
Machine-surfaced signals from a cold read, not a verified defect list. Some will prove intentional, which is why each is tied to the values behind it.
What a buyer runs into first.
The lead finding is about identity. On thirteen listings, the words on the page describe a different machine from the one being sold — almost always a close relative from the same maker, which is exactly why it reads as plausible.
Critical · Copy from another machine
A sewing and embroidery machine near the top of the range is described as a sibling model.
The description and the page body both open with the sibling's copy — its name, its touchscreen, its features. A buyer weighing the two is reading one machine's pitch on the other machine's page, with no way to tell which features belong to which.
Field
What the record holds
Verdict
Title
the model being sold
correct
Description, first sentence
the sibling model, by name
disagrees
Critical · Copy from another machine
Headings that name the wrong model, under titles that name the right one.
A computerized sewing machine whose page heading carries a neighbouring model number, on both its new and its open-box listing. A straight-stitch quilter headed as a single-needle embroidery machine. A walking-foot machine headed with the name of its own sister model — which the same page's FAQ describes as a different product.
None of this stays on the page. The storefront publishes each description in its structured product data — the block search engines and assistants read to learn what a product is — and on the live re-check that block still carried the sibling's opening line. The same pattern reaches past machines: a heat press whose description is written for an add-on platen, and an accessory whose copy is written for a different machine line.
Specifications that disagree with themselves.
Most machine listings state a key figure more than once — in the description, in the specification lines and in the page body. Where the copies disagree, whichever one a buyer or a feed happens to read decides what they believe.
Specification
What the record says
Severity
Stitch memory, a commercial embroidery machine
One field gives a tenth of what the other two give
critical
Presser-foot pressure, a heavy-duty machine
Less than half the figure in one field that the other two state
critical
Shipping weight, a steam cleaner
The pound figure right and the kilogram figure a hundred too high
critical
A presser foot in a kit
A quarter-inch hemmer described as making a half-inch hem — the line copied from the entry above it
critical
Stitch speed, an industrial machine
Two maximum speeds on one page, a fifth apart, with an asterisk never explained
high
Stitch count, a straight-stitch machine
"Yes" where a number belongs
high
Two copies of a fact that agree are redundancy. Two that disagree are a coin toss the buyer does not know they are making.
Also in this theme: stitch counts one apart between two fields, two lift weight limits both labelled for the same lift type, a warranty shorter than every sibling's, a duplicated paragraph, an HTML entity in a title, a spelling error in a specification line and a height in millimetres under a label that promises inches. Thirteen findings in all, four critical.
Open box, factory serviced or refurbished.
On a pre-owned machine the condition label is the price justification. Open box, factory serviced and refurbished are different promises, with different histories and often different warranties. Several listings make more than one at once.
Critical · Condition labels
Thirty-nine listings titled open box explain on the same page that the machine is factory serviced.
The body copy carries the factory-serviced definition word for word — a block pasted in from a template — under a title that promises something else. Elsewhere, nine listings titled refurbished carry a factory-serviced tag and a factory-serviced machine lives at a web address that says open box.
High · Condition labels
And there is no condition field at all.
Condition lives in titles, tags and body copy, none of which a feed can read as condition. Nothing downstream can separate a factory-serviced machine at well under half its list price from new stock, or apply a different policy to it.
Some of these may be right in a way the record does not show — a returned unit that was also serviced, say. That is the business's question, and it is reported as one. What the record cannot do today is tell a buyer, or a feed, which one is true.
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 sibling's description is filled in. The open-box title is filled in, and so is the factory-serviced copy beneath it. The "was" price is filled in, on five listings below the price being charged. A completeness check scores every one of those fields at a hundred percent, and every one of them gives the wrong answer.
Critical · Shipping data
Zero, and exactly one thousand pounds.
In the analysed listings the two most common shipping weights are zero and 453,592 grams — one thousand pounds to the gram. Seventy-four listings that require shipping weigh nothing, among them longarms and embroidery machines at the top of the price range. Two hundred and eight weigh exactly a thousand pounds, from industrial heads to a rotary steam press whose own page gives a weight under a tenth of that.
Beside them sit two online classes and a software unit delivered by serial number, all marked as requiring shipping. On about fourteen of the weightless machines, the listing's own copy states a real weight — the number exists; it is not in the field built to carry it.
Nothing in the record says what reads that field, and this study does not supply the conclusion. Anything that rates delivery by weight reads these machines as weightless or as a thousand pounds. Opening the shipping settings turns that from an inference into a number.
The finding that did not survive.
The automated pass raised eight findings saying that listings showed buyers photographs of other products — a vacuum and cutlery on a quilting system, die-cutting dies on a vacuum. It would have been the most striking headline in the study. It was wrong.
The pass had read each image's file name as though it were the picture. EKOM opened the photographs on thirteen listings, including at least one from each of the eight findings. Every photograph that loaded shows the product being sold. What is wrong is the name on the file — and on more than four hundred listings, the lead image's file is named for another brand's product: a sewing machine whose six correct photographs are filed as another maker's machine, an embroidery design, two vacuums and a paint cart.
Medium · Image files
The right photograph, filed under another product's name.
A shopper sees the right machine. A machine reading the page does not: the storefront publishes the first image as the picture in its structured product data and its social-share preview, file name and all. A file name is a weak signal, and it is rated accordingly. It matters here for what it reveals — the number on the end of each name matches the image's position in the gallery, which is consistent with files renamed in bulk against the wrong product list.
Withdrawing eight findings is what a cold read is for. A report that had shipped them would have told a retailer its photographs were wrong when they were right — and every other finding in it would have been doubted.
74 findings sounds like 74 problems. Most of them collapse into six 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.
New listings built by copying a sibling's page
The tell: the wrong text is almost always a close relative's — a sibling model's name in the description, the neighbouring model number on the heading. The title was changed; the heading or first paragraph was not.
One fact kept in several fields, with nothing holding them together
The tell: the disagreements are almost always between copies of the same figure in different fields, not inside one.
Condition written into the title, with a shared disclaimer block
The tell: the factory-serviced definition appears word for word on listings titled open box, and there is no condition field for it to disagree with.
Weight left at a default rather than entered
The tell: zero and exactly a thousand pounds are the two most common weights, and the listings' own copy often gives the real one.
No enforced list for product type and brand
The tell: the catalog has a clear vocabulary and the defects are departures from it — forty-eight listings with no type, nineteen industrial machines typed as household ones, the retailer's own name in the brand field.
A bulk import that re-keyed records
The tell: image files named for other products and numbered by gallery position, and — outside the machines — seven SKUs shared by two or three listings at different prices, most at addresses carrying a legacy suffix.
Rewriting thirteen pages fixes thirteen pages. Checking every new listing's heading and first paragraph against its own title before it goes live fixes those and every machine added next.
Google's product data specification requires a condition "for each used and refurbished product," and defines new as a product "in its original packaging" that "has not been opened." An open-box machine is, by that definition, not new — and in this catalog no structured field says which it is.
A sewing-machine retailer sells expertise as much as inventory. The buyer on the phone is answered by someone who knows which of two sibling machines has the larger throat, and whether a unit came back through the manufacturer. Online, the fields carry that alone — and these findings are what happens when a new machine's page starts from its sibling's and nothing reads the two side by side.
What this class of defect costs is not a missed click. It is the wrong purchase, and what follows it: a buyer who reads one machine's features on another machine's page can buy on a false premise; a condition label that disagrees with itself is the kind a return dispute turns on; a machine rated as weightless, or at a thousand pounds, is a freight cost someone absorbs.
A second clock runs underneath. Some buyers now start a comparison by asking an assistant, and an assistant reads the fields before anything else: it matches the model, checks the specification, confirms the condition. On these listings it can find the sibling's name in the description, two figures for one specification, and two conditions for one machine. 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 a buyer can choose a machine from with confidence.
1 · Apply the corrections the record already answers for itself
Rewrite the borrowed headings and opening paragraphs from the manufacturer's copy for the model actually listed. Reconcile the specifications where the record settles it. Move stated weights into the weight field, fill product type and brand from the tags that already name them, and take the shipping flag off the downloads. Approved as patterns, not listing by listing.
2 · Settle the questions that gate the rest — starting with condition
Giving every pre-owned machine one structured condition value looks like a mechanical fix, and it waits. Only the business's intake and service records know whether a unit was opened and returned, went back to the manufacturer, or was used for demonstration. Which "was" prices are meant, which duplicate SKU's price is right and what the unstated weights are come from inside the business too.
3 · Hold the line at intake
Keep ongoing intelligence where manufacturer copy meets the retailer's own listings, so the next page built from a sibling's is caught the day it is published rather than found later by a buyer comparing the two.
This is how EKOM moves a catalog from insight to impact —
and keeps every machine described as itself as the business grows.