The European Commission published results in June from 132 garments sent to an accredited laboratory in Italy under its joint compliance campaign, and 49 came back carrying fibre composition labels that did not match the cloth. The headline failure rate is not the useful number. Trade coverage of the campaign put the split at 46 percent non-compliance for items bought online against 36 percent for garments taken off a shop rail — a comparison built on 104 shop-purchased garments against just 28 bought online, thin enough that the online figure reads as a signal rather than a settled rate. It still points at where the decay concentrates: the field rots fastest in the channel where it is the only evidence a shopper has. For twenty years that cost nothing, because composition sat at the bottom of the product page and nobody sorted on it. Shoppers have started sorting on it.

The move away from synthetics is now a merchandising fact, not a sustainability posture. The Business of Fashion reported in July that health anxiety about polyester, hotter summers and fatigue with cheap synthetics have pushed shoppers across generations and political lines toward cotton, wool and linen. Cotton Incorporated’s April survey, reported by Textile World, found 41 percent of US consumers aware of microplastic pollution against 17 percent in 2017; it is a cotton trade body, so read the direction rather than the magnitude. Awareness by itself changes nothing at checkout. In that same survey of 1,022 people, 59 percent said they were likely to look for clothing made with microplastic-free fibres, and looking for something is a search behaviour, not an opinion.

Composition data is wrong for a structural reason, not a careless one. The likeliest mechanism runs backward through the supply chain: the string on the product page copies the care label, which copies the mill’s declaration, which copies the yarn supplier’s, and nobody along that chain is paid to test. Circle Economy ran more than 10,000 post-consumer garments through a Fibersort scanner for the Dutch infrastructure ministry in 2019 and found the label wrong in 41 percent of cases, with cotton-polyester blends the worst offenders. This is inherited error rather than the generated error we wrote about earlier today, and it is harder to catch because nothing about it looks new. Regulation (EU) 1007/2011 permits a three percent manufacturing tolerance between declared and analysed composition, which means “roughly cotton” has a legal boundary that most catalogues have never measured themselves against.

The trade has read the fibre shift as a sourcing brief; it is closer to a data-integrity one. Article 16 of that regulation requires fibre composition to be clearly visible to the consumer before purchase, “including in cases where the purchase is made by electronic means.” Amazon runs a standing material composition defect programme for EU and UK apparel and suppresses listings that fail it. Google’s Merchant Center specification makes the material attribute required for any product that varies by material, which quietly captures a wide slice of the category. None of that infrastructure is new; what changed is the traffic now running through it.

A filter does not grade a garment; it removes it.

Filtering turns a soft data problem into a binary one. A shopper browsing a category page never noticed that a jacket’s shell was described as “cotton blend” with no percentages; a shopper who has ticked natural fibres only never sees the jacket, and the retailer never learns why. The same arithmetic governs the retrieval layer now sitting between shopper and catalogue, because an assistant asked for a linen shirt under £80 resolves that constraint against attribute text before it ranks anything. A missing or malformed composition string does not produce a bad recommendation. It produces absence, which is the one failure mode that generates no complaint, no return and no diagnostic signal.

Being wrong in the other direction carries a published price. The FTC and the Justice Department settled with Kohl’s and Walmart in April 2022 over textile products advertised as bamboo that were rayon, along with the environmental claims stacked on top, at $2.5 million and $3 million in civil penalties. The sums were trivial against those balance sheets; the route mattered more, because the FTC reached them under its Penalty Offense Authority on the strength of warning letters sent in 2010. Brussels closed its own campaign by stopping sales on 18 of the products it tested. Composition is now the only catalogue field that can cost a retailer a filtered impression and a civil penalty in the same quarter.

The strongest case against all of this is that the migration is talk. Textile Exchange put polyester at 59 percent of a global fibre market that reached a record 132 million tonnes in 2024, with the largest single increase coming from fossil-based synthetics. Stated preference has not yet moved a tonne. For the thesis to fail, that gap has to hold: composition filtering must stay a minority behaviour that never touches enough revenue to justify a testing budget. The gap does not need to close for the cost to land, because the shopper who ticks the filter is disproportionately the one paying full price for linen, and losing them silently is worse than losing volume visibly.

The deadline is already on a calendar. The Commission’s 2025–2030 ecodesign working plan names textiles a priority category, and industry trackers of the ESPR timeline put the textiles delegated act at late 2027, with the digital product passport it introduces requiring standardised, machine-readable product data. If that lands on schedule, composition stops being a string a merchandiser typed and becomes a declaration a retailer signs. Retailers have roughly two buying seasons to decide whether they find out what their garments are made of on their own timetable or on the Commission’s. Testing a catalogue costs a lab budget this year; skipping it risks a legal position in three. The firms that end up defending that legal position will not have chosen it — they will simply have skipped the lab work that would have caught the problem first.