What a recycler can actually use

The short answer, and it is not a comfortable one

Nothing you put into a product record today reaches the person who eventually has to sort the garment. If it did, it would still not settle their question, because what a sorting line has to establish is what the object in front of it is made of, and that is a measurement rather than a lookup. A record answers by repeating a declaration made upstream. An instrument answers by reading the material.

The schemes that already collect data from producers do not ask for a record about one garment. The machines that identify material do not read one. And the trade position written by the sector that would benefit asks for three attributes, none of which helps anybody identify the item in front of them.

That is a finding about the present rather than a prediction about the future. It could change, and the last section but one sets out what would change it. What it means today is that the case for structured product data has to rest on something other than a claim about what happens to a garment after it leaves your hands, because that claim is not currently supported by anything we could read.

What actually arrives at a sorting line

Three things arrive. The object, whatever survives of its labels and whatever a machine can read off the surface. There is no fourth thing. In particular there is no lookup, because the object usually carries nothing a sorter would scan and the building usually holds nothing to scan it with.

Labels are the weak link, and they are weak in three unrelated ways. Wearers cut them out. They fade past reading. And a share of them were wrong on the day they were sewn in. A coordinated enforcement exercise across eight member states tested 132 textile items and found roughly 37 per cent of those 132 non compliant on fibre labelling, with blended fabrics the worst category inside that sample. The underlying report was not retrieved by this pass, so that figure is carried here as indicative rather than as established.

The recycling industry records the same weakness from the other end. Its own position material describes labels in post-consumer streams as frequently unreadable by the time the garment reaches somebody sorting it, which is the reason the sector's ask is written around material going in rather than around identifying material coming back. No population is given for that observation on our record, so no proportion of it appears here.

Where the label fails, the line falls back on reading the surface with near infrared light. In one published trial, 253 fabric samples were run and 73 per cent of those 253 were correctly identified at roughly one sample every two seconds. The result was uneven across fibre types inside those 253 samples: viscose was identified in effectively all of its samples, while the pure cotton and pure polyester groups were among the weakest.

The physical limits behind that number are stable and none of them is a software problem. Dark fabric reflects too little light to read. The reading reaches roughly 150 micrometres into the surface, so a laminate or a multilayer construction is read at its face rather than through it. And a blend component sitting at or below about a tenth of the mix is routinely misread as absent.

A widely cited review of textile sorting reports a headline range for how much of a post-consumer stream can be identified from label inspection alone. We do not print it. We could not obtain the number of items behind that range, and a proportion whose population nobody states is not evidence, whichever direction it happens to point.

The problem is measurement, not retrieval

The usual argument for a passport at end of life is a retrieval argument. The information existed once, the object lost it and a record hands it back. Set that beside what a sorting line is actually doing and it answers a question nobody at the line is asking.

A line establishes composition by measuring the object. A record states composition by repeating what somebody wrote down when the garment was made. Those are different operations and they fail in different ways. The instrument fails on dark fabric, on layered construction and on small blend fractions, and it fails visibly, in that a weak reading looks like a weak reading. A record fails silently, because it returns the declaration it was given and holds nothing that distinguishes a true declaration from a false one.

That distinction decides the argument, because a measured share of those declarations is wrong before anybody digitises it. Of the 132 items tested in the enforcement exercise above, roughly 37 per cent were non compliant on fibre labelling, and blends were the worst category inside those 132. A passport populated from the same declaration inherits the same error and delivers it faster. A spectrometer does not inherit it, because it never consults the declaration in the first place.

The rule about laboratories and lines

The trial figure above was produced on prepared samples under controlled conditions. Where the same kind of identification has been measured in an adjacent material stream, first under semi-industrial conditions and then under industrial ones, accuracy fell between the two. The size of that fall sits in the estate's research record rather than on this page, because this pass did not obtain the counts at their own source, and a figure without its population does not appear here.

Treat the direction as a reading rule. A trial reporting the same accuracy on a running line as it reported in a laboratory has not been run on a line. Line conditions are throughput, one presentation of each object, items folded and overlapping, damp or soiled material and trims still attached, at whatever speed the plant is paid to run. Every one of those degrades a reading and none of them is present in a sample tray.

What a working machine-reading case looks like

It is worth saying what the successful version of this looks like, because one exists and it is not in textiles. In a packaging stream, a marker applied to the object at the point of manufacture is read by detection equipment installed on the sorting line for that purpose, and the arrangement has been trialled to the point of reporting detection counts, a range of sorting efficiency and a stated duration. Those three figures sit in the estate's research record and are not published here, because this pass did not obtain them at their own source.

The contrast is the point, and the differences have to be stated because they are what make it a contrast rather than a precedent. The material is different. Rigid, largely single-material packaging carries an applied surface marker to the line intact, while a garment is flexible, blended, laundered, altered and routinely stripped of whatever was attached to it. The failure mode is different too. In packaging the problem is retrieval, and a marker plus a reader solves retrieval, because the fact the line needs is the identity of a material the marker was applied to. In textiles the problem is retrieval and measurement at once, and no marker settles the proportion by weight of the fibres inside the fabric it is stitched to. The packaging case shows what has to be true for machine reading to work. Textiles satisfy neither half of it today.

What producer responsibility schemes actually consume

The schemes are the part of this landscape that already exists, is funded and has a legal hook into your business. They are also the part that most clearly does not want an item-level record, and it is worth being precise about what they do want.

The French scheme takes an annual declaration of quantities placed on that market by product category, against a published nomenclature, with fees modulated on a small set of criteria. The Dutch scheme takes weight placed on the market together with reported reuse and material outcomes. Two further schemes we looked at work per kilogram. One member state's position we could not establish at all.

The unit of account, in other words, is the category and the kilogram. Nothing in that architecture has a place for your identifier, and a scheme reporting in tonnes has no use for a record about one garment. We did not find an operating European scheme that consumes item-level product data. That is a statement about where we looked rather than about the world, and how an absence of that kind is typed before it is written down is set out in how we know.

What the instrument actually is

The European instrument behind all of this is widely described as the textile extended producer responsibility directive, and that description misplaces it. It is an amending directive. What it amends is the waste framework directive, and the operative textile provisions are new articles of that amended directive rather than articles of the amending instrument. It entered into force in October 2025, and which products fall inside it is set by commodity codes in a new annex to the directive it amends rather than by a description of a garment.

This page publishes no article number for those provisions. The amended directive was read here for its structure and its metadata rather than for its textile articles, and citing an article number nobody in this estate has read is the exact failure this correction exists to fix in other people. When schemes have to be established is a duty rather than a property of the instrument, so it is maintained on the status tracker and not asserted here.

Who would pay for item-level data

A producer responsibility fee is the only money in this system that runs from a brand towards the people who handle the garment at the end, which makes the basis of assessment the whole answer to who pays. That fee is assessed per product placed on the market rather than per item in a sorter's hands. Money raised that way funds collection, sorting and treatment in aggregate.

Two things follow. A fee assessed on products placed on the market buys nothing about one garment, whatever the record about that garment contains. And no scheme is required to be running yet, on a timing the status tracker maintains. So there is no fee mechanism operating in Europe today that pays anybody to hold, read or act on item-level textile data, which is a more useful answer to the funding question than an argument about whether it would be nice to have.

The three attributes the recycling industry has actually asked for

There is a published industry position on what a passport should carry, written by the people who would use it. It is short, and its shortness is the useful part. It asks for three things.

• The quantity of recycled content.
• The type of waste source, meaning post-consumer, post-industrial or pre-consumer.
• The country in which end-of-waste status was reached.

Read the three together and their shape is unmistakable. Every one of them describes recycled material going into a new product. Not one of them helps identify an old one. The same position frames the passport as an instrument for enforcement and traceability rather than as an operational input to sorting, and the interest being protected is protection from competitors whose recycled content nobody can check.

That is a more useful finding than a longer wish list would have been, because it is answerable. Two of the three are attributes a brand can hold today. One of them is also the single hardest attribute in this estate to substantiate, for reasons that have nothing to do with sorting and everything to do with what a chain of custody certificate does and does not prove. This position paper reached the page through the estate's research record rather than at an address we hold, so it is described here and quoted nowhere.

Set that ask beside the legibility finding earlier on this page and the problem changes shape. The industry is not asking for more information about the object a sorter is holding. It is asking for a few checkable facts about material going in, because it has given up on the object arriving with anything readable attached. So the question circularity puts to a brand is not really what to record. It is whether anything carrying the record survives to the point where somebody would use it, and that is a question about carriers rather than about data.

What actually disqualifies a garment as feedstock

The processors are more specific than the schemes and more specific than the position papers, for the straightforward reason that they are describing a machine which either accepts an input or does not.

One cotton recycler's published feedstock specification asks for mono-fibre or cotton-rich input and names metallic yarns, rubber trims and foil prints as disqualifiers. A separate industry white paper on sorting for recycling names trims, adhesives, laminates, threads, fasteners and labels as the primary disruptors. Elastane blocks several chemical routes on its own. Both of those reached this pass through a research review rather than at their own documents, and neither is quoted here.

Nobody is currently asking a brand to record any of it. It is also the material most likely to be asked for eventually, and it is precisely the material a brand knows at the point of design and has forgotten within a season. An elastane percentage and a list of what is attached to the fabric would be worth more to a processor than the entire consumer-facing half of a passport, on one condition. It has to be readable at line speed, which hands the problem straight back to the carrier rather than to the record.

What this page will not claim

We do not say that a passport improves sorting outcomes. Three separate things stand in the way and only one of them is about data at all.

1. The record is not at the line. Sorting acts on an object whose link to any database has typically been cut out, has faded or was never readable by the equipment in that building.
2. The identification the line needs is a measurement of the material in front of it. A record asserting a composition does not become a measurement of the item by being correct about the model, and it is wrong in the same direction as the label whenever the label was wrong.
3. No outcome we found separates the effect of a product record from everything else that changed alongside it. Where a figure of that shape circulates, two questions settle it: what was the denominator and what was the counterfactual. We did not find a source that answered both.

So no recovery figure appears anywhere on this page attributed to product data, and none will until somebody produces one that survives those two questions. No scan rate appears either. There is no single rate to publish, and the measurements that do exist, with their populations and their limits, are set out at what a scan tells you.

What would change this page is narrow and specific. A scheme beginning to consume item-level data, a carrier with published survival data behind it, or a sorting study with a stated sample that supersedes the trial above would each move it. Any of them would be welcome, and the page is built to be rewritten rather than defended.

What is worth doing anyway

None of this argues for doing nothing. It argues against doing it for the reason that is usually given, which is a different and considerably more useful conclusion. Four things survive everything above.

1. Get composition right at the level the labelling rules already govern. It is the attribute the enforcement sampling above found wrong most often, it is governed now rather than later, and a wrong declaration stays wrong however well it is digitised.
2. Record what is attached to the fabric. Trims, coatings, laminates and the elastane fraction. Nobody asks for these, and the processors name them as the things that decide.
3. Record recycled content with its chain of custody model beside it. The quantity is the half nobody disputes. The model is the half that decides what you are allowed to say about the garment on the shelf.
4. Keep the record available for longer than you expect to need it. Whoever meets the garment next meets it long after you have stopped thinking about it.

Where each of those sits among the attributes we track, and how far the evidence behind each one actually reaches, is set out one row at a time. And write down what you cannot answer, with the reason you cannot. A recorded absence is usable by the next person who picks the question up. An empty cell is a question nobody knows has been asked.

Sources

• Ecodesign for Sustainable Products Regulation (EU) 2024/1781, consolidated text
  Arts. 9(2), 11CELEX 02024R1781-20240628In forceRelevant provisions reviewed

  Read at its own text and used here for three narrow points: that the contents of a passport are delegated to a product act rather than fixed by the framework, that the right to write to a record is delegated in the same way, and that a passport must stay available for at least the expected lifetime of the product. Nothing on this page rests on the annex or on the articles above twenty nine, none of which were readable.

  View official source

• Directive 2008/98/EC on waste, as amended for textile extended producer responsibility
  CELEX 02008L0098-20240718In forceOfficial source confirmed, detailed review pending

  Declared for one correction and for nothing else. The instrument commonly described as a textile producer responsibility directive is an amending directive, what it amends is this one, and the operative textile provisions are new articles of this amended directive rather than of the amending instrument. That correction was established at the amending instrument at the Official Journal in August 2026. This directive itself was read for its structure and metadata rather than for its textile articles, which is why no article number for those provisions appears anywhere on this page.

  View official source

• Cura and others, near infrared identification of textile fibres, in Recycling, volume 6 issue 1, article 11
  IndependentOfficial source confirmed, detailed review pending

  The published sorting trial behind the one fibre-identification figure on this page that carries a countable population. It reached this pass through the estate's research record rather than at the journal, so its result is stated with its sample size and none of its wording is quoted. The proposed address is constructed from the publisher's own URL pattern and was not fetched, which is recorded on the sidecar row. Neither the degradation statement nor the packaging counter-example below is drawn from this trial, and neither carries a figure.

  View official source

• Recycling industry position papers and scheme fee rules for textile sorting
  Industry positionRelevant provisions reviewed

  Position papers written by the sector that would use a passport, together with scheme fee rules, read for three things: the three attributes the industry has itself asked for, the limits on label legibility in post-consumer streams, and the fact that a producer responsibility fee is assessed per product placed on the market rather than per item. The papers carry no address on our record, so they are described here and quoted nowhere, and the legibility observation is published without a proportion because no population for it was obtained.