Methodology · PCS v2.1

How the Pimfy Catalog Score works

72 isn't an opinion. Every point is computed by a fixed rule you can check. AI selects which rules apply; it assigns no points.

What is the Pimfy Catalog Score?

A 0–100 score for a fashion catalog's data quality: the weighted average of 7 components, each verified by deterministic rules — some derived from a channel's published requirements, others Pimfy's own quality thresholds, each labeled as such in the table below. Current methodology version: PCS v2.1.

Reports cite the version they were computed with, so a score is always traceable to the exact rules that produced it.

How is the score calculated?

Each product is checked against 7 weighted components (attribute completeness 25, descriptions 20, title quality 15, images 10, seo 10, variant hygiene 10, collection consistency 10 — weights sum to 100). The catalog score is the average of per-product scores, computed over sellable products only: test items, gift cards and empty drafts are excluded first.

Component	Weight	What the rules verify
Attribute completeness	`25`	15 pts — share of the universal attributes (gender, colour, composition, care, origin and the rest) present anywhere in the listingpimfy thresholdthe same list for every product, so this part of the score is comparable across catalogs 10 pts — share of the attributes specific to the detected garment category (fashion taxonomy)pimfy thresholdnot awarded when the garment type can't be identified from the title, product type or tags counted only where the scan can observe the attribute independently — never by searching for the attribute's own name, and never twice for one underlying checkpimfy thresholdan attribute nobody can satisfy is a deduction charged for our limitation, not the catalog's
Descriptions	`20`	description existspimfy threshold length ramps toward 300+ characterspimfy threshold300 is a floor, not a target; our own published guidance is 80–150 words. Partial credit below it, because 299 characters is not the same as nothing not a near-duplicate of another product (trigram similarity ≥ 85%)pimfy thresholdboilerplate repeated across a catalog is the single most common failure we see
Title quality	`15`	length in the 15–70 character rangepimfy thresholdreadability and mobile truncation, not any channel's cap. Channel limits are scored in channel readiness, where they belong not ALL CAPSpimfy threshold no SKU code in the titlepimfy threshold contains the category keywordpimfy threshold
Images	`10`	image count ramps toward 3 or morepimfy thresholdpartial credit for one and two — a product with two photos is not a product with none alt text coverage across the imagespimfy threshold
SEO	`10`	a search snippet exists that is not merely the description's first linepimfy thresholdscored only on the snippet. Title length lives in Title quality and is deliberately not counted here a second time
Variant hygiene	`10`	sizes/colors structured as variants, not split into separate listingspimfy threshold standard option naming (Size, Color, Material…)pimfy threshold
Collection consistency	`10`	attribute vocabulary and title pattern coherent with the rest of the categorypimfy threshold

catalog score = weighted average of per-product component scores — over sellable products only (test items, gift cards and empty drafts are excluded before anything is scored, and the report shows how many).

Why isn't a channel's character limit part of the score?

Because quality and conformance are different questions, and mixing them makes the number untrustworthy over time. The PCS measures structural data quality, which does not depend on where you sell. Whether a listing satisfies a particular destination is channel readiness, reported separately.

The reason is comparability. On 27 July 2026 Amazon cut product titles from 200 characters to 75. Had that limit been inside the PCS, every score computed before that date would have quietly changed meaning, and a June report would no longer be comparable with an August one. The score you were given is the score the rules gave — and it stays that way.

One honest limitation, since this page exists to be checked: a public scan cannot read your store's meta description, so at scan time the SEO component is measured on the first 160 characters of the description. Enriched products are scored on the generated SEO field.

Why do attributes weigh more than images?

Attributes are what channels require and what buyers filter by — a missing fabric composition can block an Amazon listing outright, and per Coresight Research, AI shopping agents rely on structured, machine-readable product data to evaluate and recommend products. Descriptions and titles carry search and conversion. Images affect conversion but rarely block a listing that meets the minimum. Weights are reviewed as channels change their requirements — which is exactly why the methodology carries a version.

What role does AI play in the score?

AI determines which category's rules apply. The score itself is calculated entirely by deterministic rules. That distinction matters and we would rather state it than flatter ourselves: classification decides which attributes a product is expected to have, and attribute completeness is the heaviest component at 25 points — so a misclassification moves the score, indirectly. The AI assigns no points; it selects the rule set. Classification is keyword-based, so it is wrong in cases we have not found yet — the ones we have found are listed in the changelog below and covered by a regression suite. That is why the detected category is editable, and why correcting it recomputes the score.

What are the channel-ready scores?

Shopify-ready and Amazon-ready measure how much of that specific channel's required data is already present per product, aggregated across the catalog. A catalog can read well and still miss marketplace fields — that's why they differ from the catalog score.

The per-category requirements behind those scores are public — see the listing requirement guides for every category on Amazon apparel and Shopify.

See your own numbers

The free scan runs this exact methodology against your live catalog — no signup to start, and the report is yours either way. Scan your catalog.

Methodology changelog

A report stores the version it was computed with. Older reports keep their number and their rules — a methodology revision does not retroactively rescore anything.

PCS v2.1 2026-08-16

The attributes component only counts attributes the scan can actually observe. Six of them were scored by searching the listing for the attribute's own name, so a product only earned the point if the merchant literally wrote "pocket count" — all six measured 0% across 2,319 products. They remain part of the taxonomy and are still filled by enrichment; they no longer deduct.
Colour is counted once. Primary colour, secondary colour and standard colour map were three attributes backed by the same single check, giving one signal triple weight.
The attribute weight is split: 15 points for the universal attributes every listing is measured against, 10 for those specific to the garment category. The denominator no longer depends on whether we recognised the category — which previously meant an unrecognisable product scored two to five points higher than a correctly classified one.
A product whose garment type cannot be identified from its title, product type or tags is now reported as such instead of being quietly scored on a shorter list.
Category detection treats singular and plural as the same word. "Track Pant", "The Democratic Jean" and "Lug Chelsea Boot" were previously unclassifiable, and "Wide Leg Jean" landed in a different category than "Wide Leg Jeans".
A silhouette is no longer treated as a garment type. "Wide leg" was a trouser keyword, and being the longer match it captured every wide-leg jean — the misclassification this page has cited as a known limitation since v1.0. It is fixed; the garment noun now decides.

PCS v2.0 2026-08-16

Title length is scored once. It previously counted in both Title quality and SEO, so a single defect cost points twice and title length was worth 7.75 of 100 rather than the 3.75 the table implied.
SEO no longer re-measures the description. Its snippet was derived from the description text, double-counting a component that already has 20 points of its own.
Description and image scores ramp instead of stepping. 299 characters used to score the same as none, and two images the same as none.
Every check is labeled as a Pimfy threshold or a channel requirement.
The AI's role is stated precisely: it selects which category's rules apply; it never assigns points.

PCS v1.0 2026-07-01

First published rubric.