Skip to content
GREEN FUTURETECH.
Explore

Research libraryFrom Mood Board to Specification

Chapter 11 of 14 · Part III · Draft web edition

Rights and model risks need versioned records

Procurement must establish what may be uploaded, generated, retained and used before sensitive design work enters a model.

Sections
7
Exhibits
4
Minutes
25
Sources
19
EX 44 · Human contribution: Before processing to output use 0.8, Provider documents to contextual review 0.6

The cloth is woven from this chapter's 4 exhibits: each weft is one exhibit's own data, normalised; each warp group is a section. The probe reads the exhibit's own value where it crosses. Procedural visualisation, not a measurement.

The chapter as a draft: 7 sections, 2 exhibit wefts, one shuttle.

Lead exhibit · 44Decision mapillustrativeA model contract answers only part of the rights question
Before processing to output use →Provider documents to contextual review →Service termsInput permissionTraining useHuman contributionCultural engagementProduct rights review

Read-outService terms: Before processing to output use 0.1, Provider documents to contextual review 0.1

Method and source

GFT proposed review map. Horizontal placement moves from before processing towards intended output use. Vertical placement moves from provider documents towards contextual maker review. Coordinates are arbitrary layout positions, not measured risk. The sources support the distinctions, not the coordinates or legal clearance.

Unit: illustrative schematic positions, not scores. Status: illustrative. Sources: S21, S23, S31. Illustrative figures are teaching examples, not measured factory results.

The chapter's warpBar: share of the reading · square: an exhibit

Section 11.0 · Opening · 1 min · no exhibit of its own · one movement Go to section ↓

Overture · 1 min

A maker should release a generated design only when its permission, transformation history and technical acceptance refer to the same identifiable revision. Within GFT's Cameroon textile programme, prepared in January 2024 with source notes refreshed in 2026, the industrial objective is to move from a sector baseline where about 95 percent of cotton fibre is exported raw and about 5 percent is processed locally towards higher domestic value retention, alongside the stated national target of 50 percent local processing by 2030.[R06][R07] Capturing that value requires garment specifications that buyers and factories can verify. A software subscription, an attractive image and a successful export answer different questions. GFT proposes a versioned evidence envelope that connects input references, authorised destination, model configuration, human decisions and release conditions.[S20] Unknown permissions remain unresolved decisions, rather than becoming implied consent as files move between teams. The controls below are proposed operating practices, not an account of an installed GFT system. This chapter is not investment, legal or tax advice; applicable obligations and particular designs require jurisdiction-specific review.

¶ 01 · 3 sources

alongside the stated national target of 50 percent local processing by 2030R06, R07

R06R07S20

11.13 min · 7 paragraphs · 1 exhibit

A permission to use software differs from rights in an output

Related · chapter 147 paragraphs step it

Related exhibit 57 · chapter 14Decision mapillustrativeA stop decision is a valid pilot outcome
Unresolved to resolved evidence →Localised to release-blocking consequence →Stop: critical releas…Hold: inconclusive ev…Repair: correctable g…Repair: adverse econo…Limited expansion: bo…

Read-outStop: critical release failure: Unresolved to resolved evidence 0.2, Localised to release-blocking consequence 0.9

Method and source

GFT proposed decision map informed by NIST contextual evaluation and lifecycle-monitoring guidance. Horizontal placement moves from unresolved towards resolved evidence. Vertical placement moves from localised towards release-blocking consequence. Coordinates are arbitrary layout positions. Actual decisions require the defined gates, evidence and accountable review.

Unit: illustrative decision positions, not scores. Status: illustrative. Sources: S19, S20. Illustrative figures are teaching examples, not measured factory results.

Read this exhibit in its chapter ↗
11.1 · Introduction · paragraph 1 of 7

Adobe's Firefly FAQ, updated in May 2026 and opened for this chapter, states that for features without the beta label, users can use generated outputs in commercial projects.[S31] That statement concerns the vendor's contractual conditions for a particular service tier and feature set. It does not establish that a buyer supplied an input reference lawfully, that a generated garment graphic avoids another party's intellectual property rights, or that an output receives statutory copyright protection in the territories where the garment will be made or sold.[S31] The actual feature, model route and contract still need to be identified before any design enters a commercial tech pack.

GFT recommends separating the rights review into five distinct decisions: software permission, input permission, model-training permission, output-use conditions and the substantive infringement assessment for the intended product. Each review should name the exact object being assessed and its authorised destination. An image may be approved as an internal mood-board reference without being approved for upload to an external model, inclusion in public advertising or rotary screen printing on a garment. A file-level label such as "licensed" is therefore insufficient unless its scope, territory and expiry travel with the asset record. This is a proposed control structure, not a declaration about a particular supplier licence.

For the running cotton-shirt workflow, the proposed record would distinguish a buyer's silhouette reference, a maker's base pattern block, a surface artwork tile and the generated composite presentation. The reviewer should identify the relevant permission for each input component and then assess the proposed combination. A permission granted for internal design exploration should not silently become permission for commercial garment reproduction. When the intended activity changes from internal review to buyer sampling or bulk manufacture, the accountable owner should reopen the decision and record the reason, relevant correspondence and affected specification revisions. Retaining an earlier approval preserves history; it does not extend its scope.

Official policy documents show why a maker cannot rely on one universal rule for copyrightability or disclosure. In the United States, the Copyright Office's March 2023 registration guidance states that applicants have a duty to disclose artificial-intelligence-generated content in registration applications.[S1101] Its January 2025 Part 2 report concludes that generative outputs receive copyright protection only where a human author has determined sufficient expressive elements, distinguishing creative arrangements or modifications from supplying prompts.[S21] By contrast, the Hong Kong SAR Commerce and Economic Development Bureau noted in its July 2024 consultation announcement that under the existing Copyright Ordinance (Cap. 528), AI-generated works are already protected by copyright under computer-generated work provisions, while infringement liability and a text-and-data-mining exception were examined separately.[S1102] Neither document determines the legal position in Cameroon or an export market, and neither clears a specific shirt design.[S21][S1102] Copyrightability, disclosure, training exceptions and service permissions must remain separate entries in the decision record.

The proposed rights map in Exhibit 44 separates where evidence comes from and when it is reviewed across the specification workflow. Its coordinates only arrange the questions for reading. They are not legal scores, risk probabilities or an assurance that any question has been resolved.

A useful approval statement should name the asset revision, permitted activity, approving role, territorial restrictions and evidence location. GFT recommends distinct workflow states for permission requested, evidence received, reviewed, approved with conditions, refused and expired. An unanswered request remains pending and blocks downstream release. The workflow should let a designer continue with independently authorised alternatives while holding the affected asset. This gives uncertainty a visible consequence without treating every incomplete record as a reason to halt unrelated styles.

Procurement should also distinguish the right to use an output from any contractual defence offered by a software supplier. GFT proposes recording the actual contract clause, eligible model version, account tier, exclusions, required user filters and notification procedure for raising a concern.[S31] A marketing summary should be filed as a claim awaiting confirmation where the contract has not been inspected. No indemnity, exclusive ownership or training-data warranty is assumed in this chapter. The reviewable result is an identified agreement and a bounded decision, rather than a general declaration that a tool is commercially safe.

11.23 min · 9 paragraphs · 1 exhibit

Human contribution should be recorded without promising protection

Step 1 of 4Inputs

Exhibit 45 · illustrativeThe design record needs a provenance envelope

InputsReference IDsPermission scopeUpload decisionProcessingModel and routeConfigurationRun recordHuman decisionsSelected alternativesSubstantive editsAuthor accountAcceptanceTechnical checksRights decisionRelease revision
InputsReference IDsPermission scopeUpload decisionProcessingModel and routeConfigurationRun recordHuman decisionsSelected alternativesSubstantive editsAuthor accountAcceptanceTechnical checksRights decisionRelease revision

Inputs

Inputs: Reference IDs, Permission scope, Upload decision · illustrative

GFT proposes keeping a versioned decision history that explains what a person actually created, selected, modified and approved.

Processing

Processing: Model and route, Configuration, Run record · illustrative

Human decisions

Human decisions: Selected alternatives, Substantive edits, Author account · illustrative

Acceptance

Acceptance: Technical checks, Rights decision, Release revision · illustrative

The method behind it: Reconciling human edits, pattern states and provenance manifests.

Method and source · Exhibit 45

GFT proposed evidence structure informed by NIST provenance guidance and C2PA manifest chaining principles. Each layer links to the same identified design revision. Retention and access must be defined separately; collecting provenance does not authorise storage of restricted material. No installed system or certification is asserted.

Unit: illustrative record layers. Status: illustrative. Sources: S20, S1103. Illustrative figures are teaching examples, not measured factory results.

11.2 · the bodyExhibit 45 above; its figures are illustrative where marked.

GFT proposes keeping a versioned decision history that explains what a person actually created, selected, modified and approved. Useful evidence includes the initial hand sketch, an annotated design alternative, a revised surface repeat tile, a manually edited pattern panel and the accepted construction specification. The record should preserve the links between these objects across revisions. An export timestamp alone cannot explain whether the technician redrew a curve, chose a parametric option or accepted a machine suggestion without modification. Those actions represent different technical and creative contributions even when they produce the same final file format.

The human-contribution entry should describe visible changes in ordinary technical language. Recording that the technician redrew the two-piece collar stand curve and shifted the chest pocket placement is far more inspectable than stating that the team improved the output. The author should connect that description to before-and-after artefacts, rather than claim an unmeasured share of authorship. For the illustrative shirt, the maker can retain the chosen sketch, rejected collar options, generated program and edited CAD drawing.[S09] This is a proposed method, not a guarantee that an edit crosses a legal threshold.[S21]

¶ 02 · 2 sources

S09S21

A prompt log has a narrower purpose. It helps explain instructions, negative constraints and run behaviour, so GFT recommends keeping it alongside the expressive and geometric artefacts. Under the US Copyright Office's March 2023 guidance and January 2025 report, applicants must disclose AI-generated material and cannot rely on prompts alone for human authorship over machine-determined expression.[S1101][S21] Even where a jurisdiction such as Hong Kong SAR provides statutory rules for computer-generated works, distinguishing human-authored drawings from machine outputs remains essential for contractual disclosures and cross-border sales.[S1102] Prompt length or iteration count should never serve as a proxy for protectable authorship.

¶ 03 · 3 sources

S1101S21S1102

Recent research and technical standards show how discrete transformation records can be structured. TailorCoPilot pairs an agentic interface with TailorTrace, a version-controlled mechanism storing discrete pattern states and editing operations to support apprenticeship and auditability.[S15] In media provenance, Version 2.1 of the Coalition for Content Provenance and Authenticity (C2PA) specification, published in September 2024, defines a manifest structure where each time an asset is changed, existing provenance is preserved and the new change is added.[S1103] Neither mechanism settles third-party infringement, but both illustrate GFT's recommended principle: preserve each state transition as an explicit, linked record rather than overwriting files in place.[S15][S1103]

¶ 04 · 2 sources

Version 2.1 of the Coalition for Content Provenance and Authenticity (C2PA) specificationS1103

S15S1103

The provenance envelope in Exhibit 45 is a proposed way to connect these records without merging their meanings. It draws on the NIST generative-AI profile's attention to content provenance and lifecycle documentation.[S20] The specific layers, fields and division of responsibilities are GFT's design recommendations, not a prescribed NIST apparel system.

¶ 05 · 1 sourceExhibit 45

S20

GFT recommends that a release record point to immutable copies or controlled versions of the accepted artefacts. A later edit should create a numbered successor revision rather than overwrite the evidence behind an earlier decision.[S1103] File digests can be used to check whether the reviewed bytes match the delivered bytes. The digest should be treated as an integrity check within the proposed system, not as evidence that the content was licensed, dimensionally accurate or authored by the person named. Those questions require their own supporting records.

¶ 06 · 1 source

S1103

The evidence envelope should preserve uncertainty in contribution. Where a reviewer cannot distinguish a manually drawn element from generated material, the record should state that ambiguity explicitly.[S1101] Where a collaborator supplies only a flattened image, the maker should request the editable source or document its absence. A decision to proceed should state which evidence was missing and why the intended activity remains acceptable under the relevant review. Silence should not become an invented account of creative work.

¶ 07 · 1 source

S1101

Access to authorship evidence needs boundaries. GFT recommends a concise review copy identifying the design decisions, with restricted links to confidential source files. Not every downstream manufacturer needs access to rejected buyer references or private correspondence. The aim is to let an authorised reviewer reconstruct the decision with sufficient evidence. Retaining every intermediate file indefinitely is not assumed; the retention rule must reflect purpose, permission and approved obligations.

¶ 08

11.33 min · 8 paragraphs · read with a related exhibit

Sensitive data needs a declared destination

Related · chapter 138 paragraphs step it

Related exhibit 55 · chapter 13ScheduleillustrativePilot work should be sequenced by dependencies
0246Prerequisites and protocolMatched manual baselineAssisted case attemptsMaterial and fit checksReadback and exit rehearsalEvidence review and decision

Read-outPrerequisites and protocol: from 0 to 1 (illustrative dependency positions, not elapsed time)

Method and source

GFT proposed sequence. Prerequisites include authorised inputs, approved blocks, material records, technicians and a frozen protocol. Baseline and assisted work use matched conditions. Physical checks and receiving-CAD readback depend on suitable candidates. Numeric positions show order only; team capacity, durations and calendar dates are unknown. Review can stop, repair or bound expansion.

Unit: illustrative dependency positions, not elapsed time. Status: illustrative. Sources: S19, S20, S22. Illustrative figures are teaching examples, not measured factory results.

Read this exhibit in its chapter ↗
11.3 · Introduction · paragraph 1 of 8

The NIST generative-AI profile identifies data privacy and information security among the primary lifecycle risks.[S20] GFT's proposed response is to classify inputs before they enter a service and record the actual processing destination. The review boundary should include the selected model route and connected plug-ins, rather than stopping at the visible application's name. This is a procurement and workflow recommendation, not a finding that any named vendor exposes confidential data.[S20]

GFT recommends separate upload decisions for four apparel data classes: public references, proprietary pattern blocks, buyer designs and wearer body data. Public availability should trigger a rights check rather than automatic upload approval.[S21][S23] Proprietary blocks and grading tables should be routed only through a configuration approved for confidential geometry.[S09][S22] Buyer designs should retain their contractual restrictions. Body measurements and images should have a defined purpose, access rule and retention decision before processing.[S20][S30] These are proposed classes for the apparel workflow, not statements of applicable privacy law.

The upload authorisation should identify the receiving account, model provider, processing region where established, retention terms, training-use terms, human-access conditions and deletion mechanism.[S20][S31] An unknown field should be marked unknown. GFT recommends testing the documented behaviour with permitted synthetic material, such as open synthetic pattern specimens from a public dataset, before authorising sensitive inputs.[S30] The record should distinguish a policy claim, an observed test result and a contractual commitment. A test confirming that a visible history can be deleted should not be recorded as proof that all provider-held copies have disappeared.

HAIGEN's revised research abstract (first submitted in August 2024 and read in its July 2026 revision) describes a four-module system combining cloud inspiration generation with local sketch-library, recommendation and colouring modules, noting a privacy rationale for that split.[S07] It is a research example of separating tasks by destination. This chapter has not audited its code, network traffic or storage behaviour, and its current abstract reflects later revisions rather than the original 2024 submission.[S07] GFT should not treat the authors' privacy claim as an independently verified guarantee.

For the shirt workflow, GFT proposes sending only an approved generic description to an external ideation service while keeping a restricted base block and physical fabric records within the maker's authorised environment.[S07][S22][S32] Any generated concept or parameter file returns as an unapproved candidate that must pass validation before opening in CAD.[S09][S10] This arrangement should itself be tested: an assistant that retrieves the block or serialises coordinates through an integration changes the boundary even if the user never attaches a file manually.[S10][S20] The control should describe permitted flows and check them in the intended configuration, rather than rely on a label such as "local".

A provenance system can create its own concentration of sensitive material. GFT recommends storing identifiers, hashes and permission decisions in the ordinary design record while restricting access to body images, contracts and buyer files.[S20][S1103] Logs should omit unnecessary source content and should never become a second uncontrolled archive. The relevant reviewer needs to know where evidence can be inspected and under whose authority; other users need only the release state and conditions.

GFT also recommends an exit check before adoption. The maker should establish which approved artefacts, pattern geometries, material records and account histories can be exported in standard formats such as DXF, which formats remain usable without the service, and what deletion evidence the provider can supply.[S22][S31] Restricted information should not be copied into personal accounts to bypass an incomplete procurement review. When access ends, the design record should retain the decision trail allowed by the relevant retention rule and identify anything that could not be independently verified.

Interlude · the cloth half-woven

4 exhibits carry this chapter's argument.

Woven: sections 11.0 to 11.3Still to weave: 11.4 to 11.7

  1. 44A model contract answers only part of the rights questionDecision map
  2. 45The design record needs a provenance envelopeLayers
  3. 46A model update should reopen the relevant test gateFlow
  4. 47Procurement should demand an inspectable evidence packRegister
Hover or focus an exhibit to light its weft; select it to jump to it.
Key figures from chapter 05 · Exhibit 18Key figuressourcedAutoSew reports distinct relation and assembly metrics
  • 96.2%Total relation F1Precision 96.1%; recall 96.3%; all evaluated relations
  • 84.7%Multi-edge relation F1Precision 80.4%; recall 89.5%; multi-edge subset
  • 73.3%Error-free digital assemblyFully resolved test patterns divided by all test patterns

Read-outTotal relation F1: 96.2%. Precision 96.1%; recall 96.3%; all evaluated relations

Method and source

Sourced from AutoSew full-text sections 8.1 and 8.2. Total and multi-edge metrics concern stitch correspondences on the authors' test set. Garment sewn percentage is fully resolved test patterns divided by all test patterns. The abstract rounds total F1 to 96 percent. These are digital benchmark results, not physical sewing yield or GFT performance.

Unit: percent, with different denominators. Status: sourced. Sources: S12.

Read this exhibit in its chapter ↗

11.43 min · 7 paragraphs · read with a related exhibit

Community engagement cannot be replaced by similarity screening

The World Intellectual Property Organization (WIPO) describes traditional cultural expressions as integral to the cultural and social identities of indigenous and local communities, embodying know-how and skills and transmitting core values and beliefs.[S24] Its fashion guidance promotes dialogue between Indigenous Peoples and fashion companies to shape fair and equitable interaction and warns against harmful uses without consent.[S23] These institutional explanations support careful engagement; they do not identify a representative authorised to approve a particular Cameroonian motif, such as Grassfields ndop resist-dyed geometry. The fashion page is dated 2 November 2023; the traditional-cultural-expressions page displays no publication date. Both were reopened on 30 October 2026; neither is a new statute.[S23][S24]

¶ 01 · 2 sources

Both were reopened on 30 October 2026S23, S24

Related · Exhibit 33 · chapter 08

S24S23

GFT proposes a cultural-reference record whenever the design brief intentionally draws on an identifiable tradition or community expression. It should describe the reference, its source, the intended garment application and the proposed public account of its origin. The maker should then establish who should be consulted and what evidence supports that identification.[S23][S24] A photograph credit, a retailer's description or a model's fluent summary should not be substituted for that investigation. Where authority remains contested or unclear, the unresolved position should be visible to the release reviewer.

¶ 02 · 2 sources

S23S24

Similarity screening has a limited role in this proposed workflow. It can be used as a search aid to identify visual references for further review.[S20] GFT should not assign it authority to decide cultural permission. A low similarity result should not close an engagement question, and a high similarity result should prompt examination of the actual expression, context and intended use.[S23][S24] The review should preserve the search scope and limitations so that an incomplete search is not later described as comprehensive clearance.

¶ 03 · 3 sources

S20S23S24

For a culturally referenced shirt surface, GFT recommends linking any consultation record directly to the specific artwork revision and placement map shown during dialogue.[S23][S1103] A general discussion about a collection should not automatically approve every generated variation or colourway. The record should distinguish permission to study a reference, approval to adapt it, approval to manufacture, and agreement about public attribution.[S23] Proposed benefit arrangements, restrictions and review opportunities should be retained where relevant. These are recommended questions for a responsible process, not invented terms attributed to any community.

¶ 04 · 2 sources

S23S1103

GFT's proposed engagement record should also make room for refusal, limited permission and a request for further discussion. A design programme should not record only favourable outcomes or treat participation in a meeting as endorsement. The responsible role should document the response accurately and identify which changes (such as altering motif proportions, crossing seam joins or changing garment category) require renewed contact.[S23][S24] Reusing an approval requires matching its scope.

¶ 05 · 2 sources

S23S24

The same record should govern the language accompanying the garment. GFT recommends reviewing claims such as authentic, traditional or community-approved against the actual agreement and evidence.[S20][S23] A designer's interest in an expression is not evidence of provenance for the product. The maker should preserve the distinction between inspiration, authorised collaboration and independently verified origin. If no adequate account can be supported, the public description should be revised or the cultural reference removed through the documented review process.

¶ 06 · 2 sources

S20S23

Community engagement also needs appropriate information handling. The proposed record should capture the decision and authorised terms without placing restricted stories, images or correspondence in an unrestricted asset library.[S20][S24] GFT recommends agreeing what may be retained and shown to technical partners, then carrying those limits into subsequent generation and publishing workflows. A later supplier should receive only the material and permissions required for its role.

¶ 07 · 2 sources

S20S24

11.53 min · 6 paragraphs · 1 exhibit

Model changes require a regression decision

Step 1 of 6Identify change

Exhibit 46 · illustrativeA model update should reopen the relevant test gate

Identify changeAssess affectedgatesRun retained casesReview regressionsApprove bounded useHold and correct
Identify changeAssess affected gatesRun retained casesReview regressionsApprove bounded useHold and correct

Identify change

Identify change. Starting point. Leads to Assess affected gates. · illustrative

NIST’s Artificial Intelligence Risk Management Framework (AI RMF 1.0) notes that while the framework can be used to prioritise risk, it does not prescribe risk tolerance…

Assess affected gates

Assess affected gates. From Identify change. Leads to Run retained cases. · illustrative

Run retained cases

Run retained cases. From Assess affected gates. Leads to Review regressions. · illustrative

Review regressions

Review regressions. From Run retained cases. Leads to Approve bounded use (Conditions met), Hold and correct (Blocking issue). · illustrative

Approve bounded use

Approve bounded use. From Review regressions (Conditions met). End point. · illustrative

Hold and correct

Hold and correct. From Review regressions (Blocking issue). End point. · illustrative

The method behind it: Worked example: an improved total can conceal a blocking regression.

Method and source · Exhibit 46

GFT proposed change-control process informed by NIST evaluation guidance. A hold preserves the last authorised workflow only if it remains available and suitable. A rollback decision cannot retract an earlier disclosure. No benchmark result or deployment is claimed.

Unit: illustrative decision sequence. Status: illustrative. Sources: S19, S20. Illustrative figures are teaching examples, not measured factory results.

11.5 · the bodyExhibit 46 above; its figures are illustrative where marked.

NIST's Artificial Intelligence Risk Management Framework (AI RMF 1.0) notes that while the framework can be used to prioritise risk, it does not prescribe risk tolerance, and calls for documented evaluation under conditions similar to intended deployment.[S19] Its generative-AI profile cautions against extrapolating capabilities from narrow assessments and supports lifecycle monitoring.[S20] GFT proposes applying that principle at the boundary between design intent and accepted specification. Neither document sets an apparel pass rate or certifies the maker's workflow; local acceptance conditions must be defined before testing.[S19][S20]

¶ 01 · 2 sources

…Risk Management Framework (AI RMF 1.0) notes that while the framework can be used…S19

Related · Exhibit 08 · chapter 02

S19S20

In generative apparel pipelines, model updates can affect multimodal language models, parameter estimators, parametric pattern programs or stitch-prediction networks.[S09][S10][S12] The proposed version record should identify the model, provider route, available version identifier, prompt template, configuration, reference set and export tools.[S20][S22] When a provider does not expose an immutable version, the record should state that limitation and retain the execution timestamp and disclosed service identity. Reproducing the exact pixels is not assumed; the maker should retain the accepted output and test whether the revised workflow still meets the defined conditions for its task.

¶ 02 · 5 sources

S09S10S12S20S22

Research benchmarks illustrate why regression testing must check whole-specification outcomes alongside component metrics. In geometric stitch prediction, AutoSew reports a 96 percent F1-score on edge relations alongside a 73.3 percent error-free assembly rate on test garments.[S12] Those two figures use different denominators: high edge-level accuracy can coexist with failed garment assembly when a critical seam pair fails.[S12] GFT therefore recommends a retained evaluation set containing authorised reference blocks, known failure cases and held-out variations that test the complete handoff from permissions to CAD readback.[S19][S22][S30] Test cases should be labelled by what they challenge, including instruction compliance, missing material data, seam topology and DXF unit scaling.[S12][S22][S32] Cases used repeatedly to adjust prompts should be identified as development cases, while held-out cases remain separate until the designated evaluation.

¶ 03 · 5 sources

AutoSew reports a 96 percent F1-score on edge relations alongside a 73.3 percent error-free…S12

S12S19S22S30S32

The proposed decision path in Exhibit 46 makes that distinction operational. A change opens an impact review; the tests selected depend on the changed boundary. The record should connect findings to the candidate configuration and show why the release authority accepted, restricted or rejected its use.[S19][S20]

¶ 04 · 2 sourcesExhibit 46

S19S20

GFT recommends preserving the old and new findings side by side, including unresolved cases and reviewer disagreement.[S15][S20] A release decision should state the permitted product envelope and any extra human checks. If the supplier removes the previous version, rollback may be unavailable; the operational alternative should then be an authorised manual CAD workflow or another qualified route.[S09][S22] A changed retention policy can reopen upload permission even when garment outputs appear unchanged.[S20][S31] Functional regression and contractual change need linked but distinct decisions.

¶ 05 · 5 sources

S15S20S09S22S31

11.62 min · 6 paragraphs · 1 exhibit

Vendor due diligence must produce reviewable artefacts

CLO's exchange documentation lists Standard DXF, AAMA and ASTM options that include units, grading, notch handling and cutting or sewing lines, while its CONNECT fabric documentation requires entering weight and thickness alongside visual maps.[S22][S32] These pages establish documented interface fields, not a verified exchange between the maker's chosen software versions or laboratory-validated fabric physics. GFT therefore recommends a sample export, material metadata check and receiving-system inspection as procurement evidence.[S22][S32] This chapter defines what evidence procurement should retain and who must inspect it.

The proposed procurement pack in Exhibit 47 converts broad assurances into inspectable requests. Its reviewer labels are responsibilities, not GFT appointments or headcount. Its hold conditions are proposed gates for the intended workflow, not allegations about any supplier.

Exhibit 47RegisterillustrativeProcurement should demand an inspectable evidence pack
Evidence request
Model-specific terms
Review owner
Procurement and legal
Hold condition
Use scope unresolved

Read-outEvidence request: Model-specific terms. Review owner: Procurement and legal. Hold condition: Use scope unresolved

Open the full register (7 rows)
Model-specific termsProcurement and legalUse scope unresolved
Data-policy recordData custodianDestination unknown
Sample editable exportPattern technicianReadback fails
Material metadata schemaQuality and sourcingBasis undisclosed
System requirementsTechnical ownerEnvironment untested
Support boundariesWorkflow ownerFailure path absent
Exit and deletion termsProcurement and dataRecovery unproven
Method and source

GFT proposed procurement checklist. Actual contracts, policies, software versions and sample files must be obtained and reviewed at adoption. No current price comparison, vendor score or accepted supplier is established.

Unit: illustrative evidence requests and decisions. Status: illustrative. Sources: S22, S31, S32. Illustrative figures are teaching examples, not measured factory results.

GFT recommends retaining the supplier's response in its original dated form (including exported DXF files, contract schedules and policy snapshots) alongside the maker's assessment.[S20][S22] A demonstration recording should identify the configuration and permitted sample material. An answer that references a beta feature, a different model or another account class should remain pending for the intended use.[S31] Procurement should distinguish evidence received from evidence inspected and evidence accepted. A complete folder is not an accepted workflow if nobody with the relevant responsibility has reviewed its contents.

¶ 01 · 3 sources

S20S22S31

The technical inspection should ask whether the receiving team can open the sample, identify units, recover required annotations and edit the intended objects, while confirming that material records separate measured values from estimates.[S22][S32] GFT proposes recording discrepancies against the supplier's claimed capability and the maker's acceptance conditions. A visually convincing three-dimensional preview should not close a two-dimensional DXF editability requirement.[S10][S22] If a proprietary format is used during generation, the exit review should establish what usable industrial exports remain when access ends and whether those files have actually been reopened.[S22]

¶ 02 · 3 sources

S22S32S10

Commercial review should examine the relationship between policies, the agreed contract and operational settings. GFT recommends asking how model substitutions are disclosed, which support activities may expose uploaded material, and how an incident should be raised.[S20][S31] Pricing remains unresearched here and requires formal quotes before Chapter 12 and Chapter 13 pilot evaluations. The procurement record should not imply that a trial account's features or permissions apply to a future production arrangement.[S31]

¶ 03 · 2 sources

S20S31

Finally, the maker should assign an owner to each pending request and connect it to a decision consequence. Missing export evidence may block technical qualification while allowing a limited ideation exercise with authorised public material.[S22][S31] Unknown upload terms should block sensitive uploads regardless of a successful export demonstration.[S20] Recording those boundaries explicitly ensures that enthusiasm for a demonstration cannot expand the permission granted.

¶ 04 · 3 sources

S22S31S20

11.72 min · 7 paragraphs · read with a related exhibit

Incidents should lead to containment and correction

GFT proposes a response record for a rights concern, an unintended confidential upload, a corrupted pattern export, a misleading material statement or an escaped specification error.[S20][S22][S23] The NIST generative-AI profile includes lifecycle monitoring, incident-related actions and information-security considerations.[S20] The response design here is GFT's apparel-specific proposal; it is not a certification, regulatory acceptance or claim that such an incident has occurred in GFT operations.

The first record should preserve the reported concern, time identified, reporting route, affected style and revision identifiers, and current release state. A report should trigger an immediate trace of the provenance envelope without converting an initial allegation into an established legal finding.[S20][S1103] GFT recommends distinguishing suspected cause, confirmed observations, containment action and unresolved scope. The responsible role should identify the relevant generated outputs, human-edited descendants, specification revisions and external recipients.[S15][S1103] Where the dependency trail is incomplete, that limitation should inform the hold rather than disappear from the report.

¶ 02 · 3 sources

S20S1103S15

Note · 3 sources

S15S1103S20

Containment should match the event. The proposed workflow suspends the affected upload route, holds the disputed artwork, quarantines the corrupted DXF export or stops use of the affected specification as appropriate.[S21][S22][S23] It identifies which unaffected work can continue and why. For an unintended external disclosure, the team must establish the destination and contractual remediation process; restoring an earlier local model version cannot undo an external transmission.[S20][S31] Required notifications and legal duties need project-specific authority rather than assumptions here.

¶ 03 · 5 sources

S21S22S23S20S31

Correction must produce a new identifiable revision.[S15][S1103] GFT recommends retaining the original evidence in a restricted archive under approved retention rules, recording the corrective change and identifying the test needed to reopen the gate. Replacing a file in place leaves downstream users unable to distinguish superseded instructions from current ones. An export or seam defect requires clean receiving-system readback of the new revision, while a rights concern requires documented resolution or an independently created replacement rather than superficial re-prompting.[S12][S21][S22][S23]

¶ 04 · 6 sources

S15S1103S12S21S22S23

Learning should then return to the workflow. The maker should decide whether to add a permanent regression case, change an upload class, revise a supplier condition or refine technician training from Chapter 10.[S15][S19][S20] Repeated exceptions should be tracked in GFT's proposed risk-matrix tool so recurring control failures remain visible across styles. A small team can allocate these duties across existing roles, provided the record separates who proposes a correction from who accepts its release.

¶ 05 · 3 sources

S15S19S20

The chapter's boundary remains the accepted design and specification. Broader model operations belong to companion paper WP14 and applicable legal duties belong to companion paper WP17, connected through the same revision identifiers.[S19][S20] Within WP10, these versioned records form a prerequisite for Chapter 12's economic model and Chapter 13's pilot qualification. GFT's recommendation is practical: make the permission decision, processing route and technical release reconstructable before expanding generated design work.

¶ 06 · 2 sources

S19S20

Check the record

The chapter keeps its sources within reach.

Every knot is a source this chapter cites; select one to read its record, or filter the register by grade. Publication dates and inherited-source limits stay explicit. Read the paper's full evidence summary.

19 sources · 161 citation marks on this pageMost cited: S20, 29 times

Grade AGrade BGrade CKnot size: citations on this page

All 19 records shown

R06 · Grade C · inherited record

Cameroon Trade Hub

Inherited evidence R06

About 95 percent of cotton fibre exported raw against 5 percent processed locally; industrialisation target of 50 percent local processing.

Publication date unrecorded. Inherited source-text record. The original publication has not been independently reopened for this web edition.

Chapter source notes

R06 Cameroon Trade Hub. Publication date unrecorded; inherited source-text check.

R07 Ecofin Agency. Publication date unrecorded; inherited source-text check.

S20 National Institute of Standards and Technology. 2024-07-26.

S31 Adobe. 2026-05-13.

S1101 U.S. Copyright Office. 2023-03-16.

S21 U.S. Copyright Office. 2025-01-29.

S1102 Commerce and Economic Development Bureau, Hong Kong Special Administrative Region Government. 2024-07-08.

S23 World Intellectual Property Organization. 2023-11-02.

S09 Authors via arXiv; CVPR 2025. 2024-12-11.

S15 Authors via arXiv; UIST 2026 listed. 2026-08-26.

S1103 Coalition for Content Provenance and Authenticity. 2024-09.

S10 Authors via arXiv; CVPR 2025. 2024-12-23.

S22 CLO Virtual Fashion. 2026-07-15.

S30 Authors via arXiv; ECCV 2024. 2024-05-27.

S07 Authors via arXiv. 2024-08-01.

S32 CLO Virtual Fashion, CONNECT Support. 2025-04-17.

S24 World Intellectual Property Organization. n.d..

S19 National Institute of Standards and Technology. 2023-01-26.

S12 Authors via arXiv; WACV 2026. 2026-02-25.

Keep the thread · chapter 11 of 14

Chapter 12 picks up the thread.

Move the shuttle along the rack to preview any chapter: its sections, exhibits and reading time. Draft web edition. This research is not investment, legal or tax advice; illustrative models are not measured factory results.

The paper's rack · 14 boltsBolt length: reading time

12Part III: Adoption requires people, rights and measured economics

Economics must count accepted styles and retained physical checks

Savings exist only where the cost of review and correction is lower than the work genuinely avoided.

  • 12.1The counterfactual must describe the same accepted product
  • 12.2A complete cost model includes rejected outputs
  • 12.3A transparent example can test the arithmetic without predicting savings
  • 12.4Breakeven depends on usable volume and actual avoidance
Reading time
24 min
Exhibits
4
Sections
7
Continue to chapter 12
← Previous: chapter 10, Technicians must own the learning loopContinue the threadChapter 12 · Economics must count accepted styles and retained physical checks

Find a thread.

Type to search.