BRENNEN

Industries / Retail, hospitality and personal services / Retail and e-commerce / Thrift and resale

Sorting and grading

Sorting and grading is deciding, item by item, whether a donated or bought-in item is fit to sell, how good it is, and where it should go: the local shop, an online listing, an export bale or recycling.

7

Milestones achieved, from 1974 to 2026.

3

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1974

    A central sorting plant

    Oxfam's Wastesaver scheme, designed by the engineer Jon Vogler, started in Huddersfield, West Yorkshire, in 1974 and later sorted and recycled textiles for the charity. After a fire closed the Huddersfield site in 2011, a new plant opened in Batley in 2013. Its site manager told the Yorkshire Post in 2018 that it handled 80 to 90 tonnes of donations a week, with six trained garment sorters pulling designer pieces for Oxfam's online shop and new-looking cheap clothing for resale in Hungary, and that shops send back anything they have not sold in two weeks.

    collection.sciencemuseumgroup.org.ukyorkshirepost.co.uk

    Credit

    Jon Vogler; Oxfam

  2. 1991

    Cards get number grades

    PSA (Professional Sports Authenticator), founded in the US in 1991, grades trading cards and memorabilia on a 10-point scale and seals each item in a tamper-evident holder, so buyers can trade on the grade without seeing the item. PSA says it has certified more than 65 million items.

    LessonA grade adds value only while graders keep up; on 30 March 2021, after a surge of submissions, PSA suspended most service levels because of a backlog, and Sports Collectors Digest reported that collectors and dealers slowed their buying and selling as a result.

    psacard.comprewarcards.comsportscollectorsdigest.com

    Credit

    PSA

  3. 2019

    One grade scale for phones

    On 12 February 2019 CTIA, the US wireless industry association, published a voluntary common standard for grading the condition of used phones. It was written with device makers, carriers, insurers, logistics firms and repairers, so that a grade means the same thing when a phone is traded in, refurbished or resold. CTIA called it the first such standard to come from the whole wireless industry.

    telecomtv.com

    Credit

    CTIA

  4. 2024

    Sorters' decisions made public

    RISE Research Institutes of Sweden, the sorting firm Wargön Innovation and the charity Myrorna published version 3 of the Clothing Dataset for Second-Hand Fashion on 19 September 2024, under the CC BY 4.0 license. It holds 31,638 garments, each with front, back and brand-label photos and labels from expert sorters: condition on a 1 to 5 scale, stains, holes, smell, and the route each took (reuse, export, recycling, repair, remake or energy recovery). A further 100 garments were labeled by several sorters so that agreement between people can be measured.

    zenodo.org

    Credit

    RISE, Wargön Innovation and Myrorna, funded by Sweden's innovation agency Vinnova

  5. 2024

    Big sorters go bust

    Soex Group, one of five European firms able to collect, sort and recycle more than 80,000 tons of used textiles a year by Innovation in Textiles' account, filed for insolvency in Germany in October 2024, blaming the collapse of its markets in Eastern Europe and Asian competitors with spare capacity. In July 2025 Texaid Deutschland filed for insolvency for its collection arm and for Resales, which runs Germany's largest sorting plant at Apolda and sorted about 16,600 tons in 2022.

    LessonHand sorting pays only while buyers of the sorted grades pay more than collecting and sorting cost.

    innovationintextiles.cominnovationintextiles.com

    Credit

    Soex Group; Texaid Deutschland

  6. 2025

    Cleaning rescues some rejects

    Researchers from the University of Alberta and the University of Wisconsin–Madison sorted 4,946 items donated to two Canadian thrift charities. Of 2,271 items examined off site, 42.7% were judged substandard: still usable, but not sellable as donated. They washed, stain-treated or mended 200 of these, with success rates of 76% to 100% by kind of treatment, and of 139 returned to a store, 81 (58.3%) were confirmed sold.

    LessonSimple cleaning can turn rejects into stock, but the pilot did not count the hours spent, so whether it pays is not known.

    journals.aau.dk

    Credit

    Wing Sem Mak, Rachel McQueen and Anika Kozlowski

  7. 2026

    Brand, not wear, sets value

    Fashion for Good, a Dutch nonprofit, and the research group Circle Economy had 8,280 garments analyzed at sorting plants in the Netherlands, Spain, Lithuania and Poland. 37% had no damage and 41% only one minor flaw, yet resale value followed brand and fashion more than condition, and only 5% to 10% of what sorters receive reaches the top "cream" grade. In a repair pilot, 18 of 164 items made a positive margin after repair.

    LessonSorting by condition alone misses what buyers actually pay for.

    fashionforgood.comfashionunited.uk

    Credit

    Fashion for Good; Circle Economy; the sorters Wtorpol, Moda Re, Humana People to People Baltic and Erdotex

Watch for

  1. Cameras on Goodwill's belt

    A team led by Min Xia, an engineering professor at Western University in Ontario, is building an AI system whose cameras scan and categorize donations on a conveyor at Goodwill, Ontario Great Lakes, and signal workers when an item is missed. Donations there are sorted by hand on tables. The team had filmed 18 hours of video for training and aimed to have the system working within a year.

    Achieved whenthe system runs on Goodwill's line and someone outside the project reports how often it sends items to the right place.

    news.westernu.ca

    Claimed by

    Western University, July 2026

  2. AI sorting for resale value

    Fashion for Good reports that in Project Rewear's pilot with the sorting-technology firm Reverse.fashion, automated sorting raised productivity by 40% and moved more garments into the high-value "vintage" and "cream" grades, and estimates that this could take a mid-sized sorter's profit from zero to €6.5 million a year.

    Achieved whena sorter using the system in daily work publishes its throughput and grade shares, checked by someone outside the pilot.

    fashionforgood.com

    Claimed by

    Fashion for Good, May 2026

  3. A one-second sorting decision

    Sympany, a Dutch textile collector, leads the SORTED project with the University of Groningen and Demcon Industrial Systems, which has built a machine that opens donation bags. The goal is a line on which an AI model decides within one second whether each item should be resold, upcycled or processed into a residual product. Sympany's sorters are labeling photos to train it.

    Achieved whenthe line runs at Sympany and an outside party reports its throughput and how often it agrees with human sorters.

    rug.nl

    Claimed by

    University of Groningen, September 2026

Open

  1. 01

    Test sorting models abroad

    A public baseline model, trained on the Swedish Clothing Dataset for Second-Hand Fashion, that predicts each garment's route (reuse, export or recycling) and its stains and holes, plus a new test set of 1,000 garments photographed and labeled the same way at a sorting room outside Sweden. Together they show how much a sorting model loses when it meets another country's donations.

    The brief
    Credit · unclaimed
  2. 02

    Mend or reject, costed

    A costed trial of cleaning and mending in a thrift store's back room. Take 400 garments that sorters would reject as stained or lightly damaged and assign them at random to two groups: one is washed, stain-treated or given a quick repair such as a button or a short seam, with the minutes timed; the other goes to the rag or recycling grade as usual. Price the treated garments by the store's normal rules, follow them through one tag rotation, and report the money they bring in, less materials, per hour of work, next to the price per kilogram the rejected half fetches.

    The brief
    Credit · unclaimed

Open 01 · the brief

Test sorting models abroad

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public baseline model, trained on the Swedish Clothing Dataset for Second-Hand Fashion, that predicts each garment's route (reuse, export or recycling) and its stains and holes, plus a new test set of 1,000 garments photographed and labeled the same way at a sorting room outside Sweden. Together they show how much a sorting model loses when it meets another country's donations.

Why it is still open

The dataset comes from two Swedish organizations only. Its author's project page still described its model pages as placeholders in 2025, and no published test of a model trained on it against another site's garments was found. Fashion for Good's Project Rewear graded 8,280 garments in four countries but has not released them as a dataset.

zenodo.orgfnauman.com

Built with what exists

The dataset from Zenodo (CC BY 4.0); open image models such as SigLIP or DINOv2 with PyTorch; a no-training baseline from a model that returns probabilities for fixed-answer questions, such as Gemini Decision 1 (see the Note "A thousand questions, one pass"); Label Studio for labeling; a phone on a stand over a gridded table, as in the Swedish set.

  1. First step

    Download the dataset, train the baseline, and ask one charity sorting room for permission to photograph 100 garments alongside its sorters' decisions.

  2. Done looks like

    Baseline route accuracy and macro-F1 (an average of accuracy across the route classes) on the Swedish 100-garment set and on the new 1,000-garment set; agreement between two local sorters on 200 of the new garments; data and code published under an open license, with the scores reproduced by someone outside the team.

Open 02 · the brief

Mend or reject, costed

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A costed trial of cleaning and mending in a thrift store's back room. Take 400 garments that sorters would reject as stained or lightly damaged and assign them at random to two groups: one is washed, stain-treated or given a quick repair such as a button or a short seam, with the minutes timed; the other goes to the rag or recycling grade as usual. Price the treated garments by the store's normal rules, follow them through one tag rotation, and report the money they bring in, less materials, per hour of work, next to the price per kilogram the rejected half fetches.

Why it is still open

The University of Alberta's pilot found that 58.3% of 139 treated items sold, but it did not time the work or keep an untreated comparison group. Fashion for Good's Project Rewear found that only 18 of 164 items made a positive margin after paid repair. No published trial compares the value of an hour spent cleaning donated clothes with what the same clothes fetch as rags.

journals.aau.dkfashionforgood.com

Built with what exists

Household laundry products, such as the oxygen-based stain remover the Alberta team used most; a sewing kit; a timer and a spreadsheet; the store's own till records of sales by tag; Python with pandas. The team handles no money; the store sells as usual.

  1. First step

    Ask one store to set aside 40 rejected garments, time the treatment of 20, and tag them so the till records their sale.

  2. Done looks like

    At least 400 garments from at least two stores, with minutes, materials cost, price and whether it sold within one rotation for every treated item; the net return per hour of work with a confidence interval; data published openly, with a second person rechecking 10% of items.