# Public apparel passport observations: methodology

**Publisher: PassportCraft. Observation date: 18 November 2026. Expanded revision 2.** We examined what a visitor could find in public garment passports and related product journeys without signing in. This is a comparison of visible information and access, not a compliance audit or a test of whether brands' claims are true.

The study used browser automation and AI-assisted observation and coding. It was not a human usability study. The original protocol preceded task coding. Each new source had a separate selection freeze, and boundary clarifications were documented during collection. The work was not formally preregistered.

## The sample

The current comparison covers **37 garment models across 15 brand or retailer programmes and 9 reader families / implementation groups**. The complete register retains **45 selected attempts**, including readers that did not load and records that do not belong in the current comparison.

| Status | Attempts | Meaning |
| --- | ---: | --- |
| `current_linked` | 37 | Exact current-linked public readers; these are the feature-count denominator. |
| `launch_linked_example` | 2 | Two explicitly labelled Lindex launch examples; outside the current cohort. |
| `adjacent_traceability` | 2 | Two Fabienne Chapot journeys whose exact website interface retains its original adjacent classification. |
| `missing_passport_entry` | 2 | Selected Artknit plain-stitch and Dedicated Stockholm entries without an established visible model reader. |
| `missing_adjacent_entry` | 1 | The selected Fabienne Chapot Philia entry without an established visible model journey. |
| `failed_reader` | 1 | Another Tomorrow’s known Seamed Waist Jacket reader did not display garment content in two anonymous sessions. |

Only records with `status = current_linked` enter feature counts. Their `included_in_current_cohort` flag is true. Examples, adjacent traceability, missing entries and failed known readers remain separate; they are not silently counted as empty current passports.

Revision 1 contained 25 current models across nine brands and six families, with 31 attempts in total. Revision 2 adds the separately documented expansion. All original 31 rows, including their notes, classifications and adjudicated task codes, remain unchanged. The earlier ten-record pilot is not pooled into either revision.

A reader family groups a shared implementation. It is not a count of independent vendors or provider tests. Lowes and its Manhattan private label form one retailer programme in the brand count. The selected chinos' Manhattan label remains explicit in the record.

## What we selected

Discovery used public search, brand product pages, first-party deployment pages and provider or project sources. Search snippets and announcements were leads; they did not establish a reader's content. A selected current record needed an exact public reader linked from a current first-party product or deployment source, a named real garment and no explicit historical-only or demo status. First-party deployment evidence could come from the brand or an implementation provider describing its own work. An attributed embedded journey could qualify. A live URL alone did not establish current deployment.

One garment model counts once. Size, colour, batch and serial variants, and duplicate resolvers, do not create additional observations. Selection allowed at most three models per brand and ten per implementation group. The original protocol specified normalized URL order within discovered frames, and selections were frozen before task scoring. Two expansion frames used documented exceptions: Lowes deliberately covered a polo, chinos and a T-shirt, and King Louie selected three distinct dress cuts. Both choices were frozen before task outcomes were inspected. They were purposive garment-type choices, not the first three products alphabetically across their catalogues. New families took priority. Additional eligible records stayed as reserves; selected failures were retained rather than replaced.

The expansion added selected attempts from Dedicated, KnowledgeCotton Apparel, King Louie, Lowes, Xandres, Weartek and Another Tomorrow. Six contribute current readable records; Another Tomorrow remains a failed known reader. The new Xandres presentation combines Tex.tracer and Twintag and is grouped separately from the original Tex.tracer modal. Their shared component means the nine implementation groups are not nine independent vendors. Discovery logs also retain unsuccessful EON/Aura/TrusTrace and other brand leads; announcements or physical-label access descriptions alone did not add records.

This is a purposive sample of discoverable public pages, shaped by English-accessible sources. It is not a random sample or an exhaustive directory. A current shop link does not prove stock, current physical-label use, service availability or legal DPP status. Unverified examples and historic pilots were not used to reach the numerical target.

## How the walkthrough worked

Collection used anonymous Chromium, an en-GB locale and a 1280 × 900 desktop viewport. The collector opened the entry page, followed the observed reader route and inspected visible sections, tabs and disclosures after rendering settled. Optional cookies were rejected where offered. Optional consent was not accepted to complete a task. Newsletter overlays could be closed without submitting information.

Each task allowed at most two relevant public destinations: the most directly labelled link, followed by a relevant alternative when needed. Shared destinations could be inspected once for an implementation group. Multiple records linking the same service are not independent service tests.

Checks stopped before purchases, ownership changes, bookings, return requests, contact messages or other transactions. Public product-number lookup was allowed when the exact model number was already published by the brand. No account, private customer data, guessed serial or batch identifier, or protected endpoint was used.

A failed page or destination received one fresh anonymous retry. We recorded HTTP status, the final address and what actually rendered separately: a 200 response can show the wrong page, while a 403 response can accompany readable garment content. A locator timeout or other automation problem is not by itself a reader defect.

Where documents were opened, they were checked for visible content and stated scope, including rendered PDF pages where needed. Not every linked file was opened; link-only observations and uninspected destinations remain explicit in the rows. Downloading a file and rendering it in the browser are different outcomes. Mobile checks used a 390 × 844 viewport. The original collection checked at least one record per included family. Expansion collectors checked the new Lowes, Xandres and Weartek implementations, plus selected KnowledgeCotton Apparel and King Louie readers. This provides at least one primary viewport check for each current implementation group, not coverage of every record, device or task. The independent expansion coder did not repeat mobile checks.

## What counts as the reader

For an embedded passport, we assessed its labelled component or modal. The parent product page could establish the garment's identity, but shop composition, care, language selectors and service links outside the component did not earn reader credit. The `host_page_context` field preserves important distinctions.

**Lowes uses a different access pattern.** Its first-party page explicitly calls the product-number search a Digital Product Passport lookup. Entering each selected published SKU led to the ordinary product page, with a product-information anchor. For those three records, the directed Description / More Information and Care Instructions tabs are the reader boundary. Global shop navigation and services remain excluded. We did not scan a physical label or establish that a physical QR resolves to the same URL. This documented boundary does not bring unrelated host-shop sections into other brands' embedded-reader results.

## Eight questions and six codes

The task columns cover product identity, named materials, quantitative composition, care, repair, supporting evidence, end-of-life guidance and language options. Read each code with its notes and access qualifications. The codes are categories; do not add them into a score.

| Code | Meaning |
| --- | --- |
| D | An answer or actionable instructions appear directly inside the assessed reader. Identity may use the named parent product. Concrete shared care advice can qualify. D does not establish independent validation or guarantee the instructions are specific to that model. |
| L | A relevant link is present. This does not mean the destination opened, the service was applicable or a transaction was completed. |
| G | General advice or context is present. For evidence, this may be a supplier narrative or brand policy without an established supporting-document route. |
| NF | Not found during the bounded walkthrough. The information may exist elsewhere or behind protected access. |
| R | Restricted access. Restrictions at an outgoing destination can instead be recorded in the notes while an exposed link retains L. |
| NA | Not assessable because no exact reader or sufficient visible content was established. This does not mean not applicable, and is never counted as NF. |

Named materials and percentages are separate questions. A supplier's description of fibres it handles does not establish the selected garment's composition. Percentages establish a visible disclosure, not its accuracy or completeness.

Repair needs an explicitly relevant instruction or service route. Generic contact alone is insufficient. General repair intentions remain G; country, audience and product-family limits stay qualified. A marketplace for buyers is not automatically an owner resale intake. Ordinary retail returns are not end-of-life take-back, and a footprint stage labelled end of life is not disposal guidance. When mixed content appears, the code reflects the most specific qualifying answer and the notes retain the rest. Generic local donation advice is G; a relevant named resale link supports L; concrete programme collection instructions can support D. Generic advice does not become D merely because it is written as an instruction.

`evidence_scope` separates model or product-family information, supplier/material declarations, certification programmes, brand documents and general methodology. A badge, filename, Approved label or accessible PDF does not establish an authenticated certificate, current validity or substantiation of every garment claim. An educational Woolmark story is not a laboratory report on the selected garment.

## Qualifications that matter

| Records | Observed qualification |
| --- | --- |
| IQONIQ | The reader rendered. A linked blockchain artifact showed a block initially, with no recorded popup HTTP status; the fresh retry returned 403. Evidence remains L, with restricted destination access. |
| Chinti & Parker; Nobody's Child | Main navigation returned 403 while named garment content and controls rendered. These are readable readers, not blocked-reader counts. Their repair links do not establish eligibility, geography or completed repair. |
| Lindex examples | Pre-Loved is linked with Sweden/Finland availability stated. The destination returned 403 twice. Product applicability and ordinary-user access were not established. These explicitly labelled examples stay outside the current cohort. |
| Fabienne Chapot Nadia/Poppie | The named MENDED link redirected to a page that rendered 404 on both checks. The repair link remains L; it does not establish service access. These records retain their original adjacent-traceability classification. |
| by-bar | Lower repair cards were not visibly established, so repair remains NA on all three records. Evidence is G for visible supplier context. Begum end of life remains NA; Benji and Bobi retain G. Separately opened policies and repair pages do not complete an unconfirmed reader route. |
| Artknit plain-stitch; Fabienne Chapot Philia | Product pages rendered, but exact model readers were not established. All eight tasks remain NA, with no reader URL. |
| Lowes | The DPP lookup opened existing product-information tabs. A delayed newsletter overlay was closed. The Manhattan private label is included within this one retailer programme. |
| Another Tomorrow | The current transparency page linked a named Seamed Waist Jacket reader. Both anonymous attempts returned HTTP 200 but displayed only the shop shell, with no garment content. All eight tasks are NA. The brand's newer platform announcement is not substituted for the observed legacy route. |
| Dedicated Stockholm | The product page and traceability accordion rendered, but the expected passport launcher was empty on the initial and fresh checks. It is retained as a missing entry. The separately preselected Malmoe sweatshirt did not replace it. |
| New Dedicated, KnowledgeCotton Apparel and King Louie readers | Supplier-profile certificate controls were visible. Evidence is L for those exposed paths; outgoing certificate files were not independently inspected or authenticated in the expansion collection. Their presence does not establish coverage of the selected garment. |
| Xandres | Direct store-based repair instructions support D. Preloved links and public Sell guidance rendered, with condition, age, label and quantity limits and vouchers stated. No appointment, account or transaction was completed. Generic local donation advice is not treated as D. |
| Weartek | Main navigation returned 403 while the named reader rendered. Care and employer collection steps appeared in images. The page discusses prototype components and testing limits; current-linked status does not establish full production or physical labels. It explicitly distinguishes fabric testing from garment testing. |

Dondup's initial proof-preview clipping was not reproduced after a settled follow-up, so it is not reported as a confirmed defect. Nobody's Child's playbook downloaded successfully even though its original browser tab stayed blank. Further qualifications remain in each row; an observed access problem is not proof that the service fails for all visitors or physical tags.

## Independent coding and adjudication

The original second AI reader coded nine records before seeing the initial codes: Dondup Amber, Artknit Merino, FUSION TWO, IQONIQ Bryce, Chinti wideleg, Nobody's Child Bailey, Finisterre Barents, Lindex bra and by-bar Begum. This covered eight current records plus one example, across nine brands and seven families. It also checked the Artknit missing-entry exception.

That reader revisited at least one entry in every original family, but did not independently repeat outgoing destinations or mobile checks. AWARE Care and Lindex accordion retries encountered locator timeouts; those codes relied on the first collection's captures. This is not a statistical inter-rater reliability estimate.

Original disagreements were preserved before adjudication. They concerned visible links versus restricted destinations, actionable shared care versus general advice, supplier context versus supporting documents, and lower-card visibility. The final decisions retained L for IQONIQ proof and Lindex Pre-Loved links with restrictions; concrete shared care remained D; Chinti's educational story remained L; Finisterre's supplier narratives remained G; by-bar's unresolved repair cards remained NA. Follow-ups confirmed Birkin's explicit organic-cotton wording and resolved Dondup's preview uncertainty. Revision 2 does not overwrite those decisions or preliminary records.

For revision 2, a second AI reader independently coded four additional records before seeing the primary codes: King Louie Phoebe, Xandres Patent leather-look skirt, Lowes Essential Polo and Weartek Enforcer. These cover four implementation groups and 32 task cells. The lead supplied target URLs, relevant controls and the Lowes boundary; source recruitment was not independently repeated. The blind file was preserved separately from adjudication.

Three initial differences were resolved without changing the primary codes. Phoebe evidence remains L because another inspected supplier profile displayed certificate controls, despite the two profiles opened by the second reader having none. Xandres end of life remains L: its named Preloved route qualifies, while local donation advice is general. Weartek end of life remains D because a displayed process image gives employee-return, employer-storage and collection steps; the second reader had not inspected that image. The lead visually checked the relevant evidence before freezing the decisions.

The second reader did not repeat mobile checks, all outgoing documents or service paths. These 32 comparisons are a documented check and adjudication exercise, not a human study or a statistical reliability estimate. The final data files were checked against all 32 adjudicated cells, and the blind file hash was checked against its recorded freeze.

## Files, calculations and reuse

`records.json` and `records.csv` contain the same 45 attempted rows and the same 29 fields per row. CSV uses UTF-8, one header, lowercase boolean flags, empty cells for null reader URLs and JSON text for the nested `supporting_sources` array. An empty source array means no source is listed, not that no evidence exists elsewhere.

`comparison-2026-11-18.csv` contains 15 current-cohort brand-group rows. Composition and care columns show D counts. Repair and evidence each include all six D/L/G/NF/R/NA categories, including zeros, so every partition reconciles with the brand denominator. A zero in a D column is not proof of universal absence.

`findings-2026-11-18.json` provides all eight task partitions, definitions, dated claims, exact matching record IDs, full denominator IDs, qualifications and source-file hashes. The aggregates are recomputed from `status = current_linked`, not from a mixture of current readers, examples and failures. The schema version is 1.1; record rows keep the original 29-field shape. The original 31 records were compared field-for-field with revision 1, and CSV was decoded back to the same typed JSON values.

Private collection captures, documents, browser logs and preliminary coding are not included in the public downloads. Exact source URLs remain factual provenance in the data; the article does not need a promotional hyperlink for every implementation provider.

In this current cohort, **23 of 37** models displayed fibre percentages and **20 of 37** displayed concrete care instructions. **Seven of 37**, across three brand groups, displayed concrete repair instructions or linked a relevant named repair route: one D and six L. General repair context, not-found and unassessable outcomes remain separate. The 29 evidence-L records include educational, brand and supplier material; they are not a count of garment certificates or authenticated claims.

These observations are dated and links may change. Shared templates and documents make rows dependent. The sample does not support industry percentages, confidence intervals, provider rankings or conclusions about protected information. We did not test service completion, claim truth, legal sufficiency or certification authenticity. The study also did not measure search demand, rankings, assistant citations or causal visibility gains.

Suggested citation: **PassportCraft. “Fashion Digital Product Passports: Lessons from 37 Examples.” Observed 18 November 2026, expanded revision 2.** [Read the article](https://passportcraft.com/insights/digital-product-passport-fashion-examples). Include the revision, relevant record or finding IDs and your access date. When reporting a count, state its status and denominator.

You may reuse and adapt PassportCraft's observation data and methodology with attribution to PassportCraft and this revision. Preserve dates, statuses and material qualifications, and identify your changes. This permission covers our observations and writing; it does not grant rights to third-party text, trademarks, images, certificates or linked material. No third-party document copies are distributed here.
