The Notebook · Essay 01
The garment as a data record
What happens when every piece of clothing has to carry a structured attribute sheet — and what we’ve learned building the machine that reads one.
A regulatory shift is underway in Europe that most people outside compliance teams haven’t noticed. For apparel data, it is one of the more consequential things to happen in a decade. Under the EU’s Ecodesign for Sustainable Products Regulation, textiles are among the first product categories slated to require a Digital Product Passport. The passport is a machine-readable record that travels with the garment and describes what it is: fibre composition, origin, care, durability, recyclability, and more. The passport is publicly reported, legislated, and coming. This is not a forecast. It’s a rulebook.
We build attribute-extraction systems for a living, so we read that rulebook differently than most. To us, the Digital Product Passport is the moment the industry formally admits something we’ve operated on for years: a garment is a data record. It always was. The regulation just makes the record mandatory, structured, and legible to a machine.
- Silhouette
- single-breasted coat
- Lapel
- notch
- Closure
- button, front
- Fibre
- wool blend
- Weight
- heavy / cold-weather
- Layer
- outerwear
The catalog was already a database — it just didn’t know it
Underneath the photography and the copy, every product page is a set of attributes. Silhouette, neckline, sleeve length, closure, fibre, weight, intended layer. A human reads the photo and infers all of it in a glance. A database needs those attributes written down explicitly, in consistent language, in fields that mean the same thing on the ten-thousandth product as they did on the first.
A person sees what a garment obviously is. A machine only knows what the garment formally states. That gap is the entire problem of our field. Most catalogs describe a garment in prose written to sell it, not in structure built to query it. The Digital Product Passport pushes the industry across that gap by force of law: the attributes stop being marketing and become the record.
What we’ve learned extracting attributes at scale
Three things, from building the extraction engine:
- The hard part isn’t reading one garment — it’s agreeing what the fields mean. Extracting “this is a coat” is comparatively easy. Where does a heavy overshirt stop being a shirt and start being outerwear? Deciding that once, and applying the same ruling to every record, is the real work. A passport is only as useful as the taxonomy behind it. Without a shared vocabulary, you get a million records that each describe a garment and collectively describe nothing.
- Structure beats prose, but prose is where the truth hides. The richest signal often lives in the description a human wrote to sell the piece. Turning that into a clean attribute means reading intent, not keywords. A jacket described as “cropped utility” has to land in the same structured record as one described in completely different words. That’s an interpretation problem, and it’s why we treat it as one.
- Once it’s a record, the catalog becomes a database you can actually ask questions of. This is the payoff. When every garment carries structured attributes, a catalog stops being a wall of pictures and becomes a queryable object — countable, comparable, joinable. The passport regulation will hand the whole industry a version of the thing we built the engine to produce.
Why this matters beyond compliance
Most coverage of the Digital Product Passport frames it as a sustainability and traceability measure, which it is. But the quieter effect is that it standardizes the shape of garment data across an entire market. When the record is mandatory and structured, the catalog becomes machine-legible by default — and the distance between “a photo of a coat” and “a row in a database” finally closes.
We’ve been on this side of the gap for years, on our own. It will be good to have company.