Resolution Layer Case Study · Read & Diagnose
Gemstone rings.
A catalog read across the gemstone rings of a colored-gemstone jewelry retailer — where every ring is described three times on its own page, and the structured specification is the copy most likely to disagree.
This is a real EKOM catalog analysis, with the client's identity removed. The client is a jewelry retailer that specializes in colored gemstones, selling to shoppers who care about exactly what they are buying — the stone, its treatment, its origin, its weight and the metal it is set in. EKOM read every ring in a set of the retailer's gemstone ring categories from its public product pages, and flagged 71 items. Eleven were withdrawn because checking showed the record was right or the finding had misread it; ten duplicated a finding already counted; six were held back because they could not be confirmed. EKOM's analyst added three the automated pass did not raise. 47 are reported here as actionable catalog defects. Nothing withdrawn, duplicated or held is in any count below.
What makes this category different is how much a ring's page says about it. The product name states the stone, the metal and a carat weight. A written description says it again. Then structured specification panels say it a third time, stone by stone: gemstone, species, shape, cut, color, composition, treatment, origin, count and weight. The first two are written for a person. The third is what a filter, a search index or any system that describes the ring without looking at it would read.
Here a wrong field can misstate the gold content of the ring, or the stone in it — while, on most of the rings in this study, the name and the description on the same page have it right.
Product codes, prices, brand and designer names, the retailer's own field labels and exact measured values have been removed or generalized; the extents (208, 288 of 552, 55, 75, 637) are the analyst's recount across every ring read; other counts are the rings a finding names, and are floors.
What's inside
- At a glance — the severity split, where the findings concentrate, and how the read was done.
- What a shopper runs into first — one ring, three descriptions, and a karat that does not match.
- The receipts — measurements in the wrong unit, stone records that contradict the stone, and a reference price stored as words.
- What held — the checks that could have failed and did not.
- Where it traces back, and what's next — the five patterns most of these findings point to, 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 pages, and the headline findings were re-checked live two weeks later.
47
Actionable findings
(71 flagged, 3 analyst-added)
Where the 47 concentrate
Theme
What it is
Findings
Name vs. record
The wrong karat, gold color or stone color; a missing stone
10
Units and measurements
Inch figures labelled millimetres; lost decimals
10
Stone records
A diamond "not faceted"; shapes their dimensions rule out
6
Stone weights and counts
Blank weights, counts and cuts
6
Controlled vocabularies
One idea under two labels
6
Treatment, formatting, navigation
"Unknown" treatments; stray commas; missing sizes
8
The reference price
The basis of a savings claim, stored only as words
1
How EKOM read this catalog
1
Read
The retailer's own public product pages — no file handoff, no credentials, no integration.
Every ring in the categories read; each category's count matched the total its own listing page declares.
→
2
Profile
Infer what each field is for from how the catalog itself uses it, then hold every ring to that.
Each stone group read in its own panel, so an accent stone is never mistaken for the center stone.
→
3
Analyze
Surface defects, rate severity, and bind each one to the rings and values behind it.
47 findings, 9 themes, 5 patterns.
Findings from an automated review of public pages, checked by an analyst — not a merchant-verified defect list. Some will prove intentional, which is why each is tied to the values behind it.
What a shopper runs into first.
The lead finding is about agreement. Each ring is described three times, and on the rings below two of the three agree — usually the two written for a person.
Critical · Name vs. record
A ring named and described as 10-karat gold is specified as 14-karat.
The product name says 10K. The written description says the ring is set in 10k gold. The structured material field says 14K. Two more rings show the same shape — one described in 18-karat gold over silver and specified at 14, one described in white gold and specified as yellow.
Where
What it says
Verdict
Product name
10-karat white gold
agrees
Written description
set in 10k white gold
agrees
Material field
14-karat white gold
disagrees
Wherever a metal filter reads that field, a shopper looking for 10-karat misses the ring, and one looking for 14-karat finds a ring its own page calls 10.
Critical · Name vs. record
A black sapphire, specified as white.
Name and description say black sapphire; the stone-color field says white. Elsewhere a ring's name promises pink and white sapphires while its record carries only the white ones, and a blue lab-created spinel is named as the aquamarine it imitates — though its own description calls it a simulant, as a sibling ring of the same stone is named. On that ring the name is the outlier; the specification records spinel.
Most of these are not missing values — each copy is filled in, and the copies disagree. The pattern is consistent with three descriptions of one ring being written separately, with nothing bringing them back into agreement.
A band thinner than paper.
Many rings state their band width twice, in two different specification panels — once in inches and once in millimetres. On the rings that state it twice, the two disagree more often than they agree.
Critical · Units and measurements
208 rings publish the inch figure with "mm" attached.
Both fields carry the same number; one is labelled inches, the other millimetres. A tenth of an inch is a normal band. The smallest of those millimetre figures are thinner than a sheet of paper. Of 552 rings that state band width in both places, 288 disagree with themselves — including those 208, and 55 more where the two figures differ in size.
Field
What it holds
Verdict
First band-width field
about a tenth of an inch
plausible
Second band-width field
the same number, in millimetres
impossible
Critical · Units and measurements
A ring top over five feet wide.
One ring's top measures a little under an inch long and more than sixty inches wide. Its own written description gives the width correctly, a fraction of an inch — the decimal appears to have been lost only in the structured field. Two more ring tops carry what look like the same slip. One designer line gives stone dimensions in inches, and on one of its rings the figure sits under a millimetre label — a factor of 25.4 wherever a filter reads the number.
None of these values survives a range check, and nothing appears to apply one. Each was still on the live page when re-checked.
Stone records, and a price stored as words.
A specification this rich has a cost: every field is one more place for a ring to disagree with itself.
Critical · Stone records
A heart-shaped diamond, recorded as "not faceted".
That value belongs to cabochons and carvings. Nearby, a gemstone name sits in a cut field, colorless mined diamonds are recorded as heat-treated, and 75 rings give a stone a shape its own dimensions rule out — a round stone measuring longer than it is wide.
Ring
What the stone record holds
Verdict
An onyx ring
Tiny ruby accents filed as the primary stone; the large onyx center as secondary — so the ring sits in the ruby category
primary is not the featured stone
An amethyst halo ring
The ring's total carat weight copied into both the amethyst and the halo stones
counted twice
Critical · The reference price
637 reference prices carry their basis only as words typed in front of the number.
Most struck-through reference prices are a bare figure. The rest name a different basis — an estimated retail value, a competitor comparison, a manufacturer's list price. On the page this works: the label is shown, with an explanation. In the data it does not. The basis lives only as text inside the price, so any system reading the field as a number either fails on it or treats every basis as the same kind of figure. The savings arithmetic itself is right on every ring read; this study makes no claim about how any reference price was set.
A study that only lists defects is not measuring anything. Four cross-checks that could have failed did not — and they say as much about this catalog as the findings do.
Check
What was tested
Result
Species
Every gemstone against its mineral species, across 49 distinct gemstones
no violations
Category
Every ring's primary stone against the gemstone category it is listed in
no violations
Savings
Reference price minus selling price against the saving shown
no mismatches
Composition
Gemstone and material composition within each stone group
agree, bar composite turquoise — a real trade distinction
Where the retailer chose from a list, the list held. The failures concentrate in two places: units, and fields that carry free text. Four stone fields use more than one label for one idea — two words for earth-mined, two for lab-made, two for untreated, and treatments packed together into single values — so any filter built on them splits one group into two. That is a far more tractable problem than a broken taxonomy.
What the study withdrew
Eleven findings did not survive checking. One group looked like carat weights that contradicted the name — until the stone records showed the name's weight was the lab-grown diamonds' weight, matching the diamond tier exactly. Others named a stone the written description confirms, misread which field held which value, reported a value the record does not hold, or rested on a false premise, and one was an artifact of EKOM's own processing. A withdrawal is the method working: a finding the record does not support does not ship.
47 findings sounds like 47 problems. Most of them point to five upstream patterns — and a pattern can be fixed once and stopped from recurring. Each tell below is a pattern in the data; the cause is EKOM's reading of it.
Three descriptions, written separately
The tell: the name and the written description agree, and the specification is the outlier — or, on the simulant, the reverse. It produced the karat, gold-color and stone-color mismatches and the missing stone.
Measurements typed as text, unit included
The tell: the unit lives inside the value, one measurement has two fields, and impossible values reach the live page. It produced the band widths in the wrong unit and the lost decimals.
Stone groups filled by position, not by role
The tell: the primary stone is not always the featured stone, a total lands in two groups at once, and a gemstone name lands in a cut field.
Free text where a list belongs
The tell: one idea under several spellings, and a qualifier typed into a value rather than stored beside it — the reference-price basis, the treatment vocabulary, trailing commas.
Some lines entered with fewer fields
The tell: the gaps cluster by product line and by stone — one designer line in inches, pre-owned rings whose only size sits in the product code, weights missing on about one sapphire in ten and almost no zircons.
Correcting three karats by hand fixes three rings. Checking the specification against the name and description at entry fixes those three and every ring added next — and the correct value is usually already written on the page.
A colored-gemstone retailer sells information as much as jewelry. Its shopper buys on the karat, the stone, its treatment and its weight, and this retailer publishes that detail in structured form, stone group by stone group. That depth is the asset. It is also why these defects matter: the specification is the part of the page a machine reader would rely on.
What this class of defect costs is not a sale — it is the wrong sale, and the trust that follows it. Wherever a filter reads these fields, a shopper filtering by karat is shown a ring its own page contradicts, a filter by stone color returns a black stone among the white ones, and a band-width filter has to discard hundreds of rings to make sense.
A second clock runs underneath. Comparison tools, marketplaces and assistants can describe a ring from its structured fields alone, without the photography or the paragraph of copy. A system that reads only the fields takes the karat, the stone color, the carat weight and the band width as given. Where the specification disagrees with the page, it carries the specification's version.
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 shopper, and a machine, can buy from with confidence.
1 · Apply the corrections the page already answers for itself
Put the band widths in the unit only one reading allows. Carry the name and description into the specification where the two agree and it does not. Split each labelled reference price into a basis and a number. Strip the stray commas, sending the two-tone rings for a decision. Approved as patterns, not ring by ring.
2 · Settle the questions that gate the rest — starting with carat weight in the name
On some rings the name's carat weight is the whole ring; on others it is the accents alone. Making them consistent looks mechanical, and it waits, because it changes what a shopper compares. So do which band-width field is authoritative, how simulants are named, whether two labels for mined stones are deliberate, and what a supplier's treatment records say. Those answers come from inside the business first.
3 · Hold the line at intake
Keep ongoing intelligence where the specification meets the name and the copy, so the next ring whose panel disagrees with its page is caught as it lands, rather than found later by a shopper filtering for it.
This is how EKOM moves a catalog from insight to impact —
and keeps every specification reading right as the business grows.