Research libraryMeasure Before You Model
Chapter 13 of 14 · Part IV · Draft web edition
The factory should earn each digital step through reliable records, active management and verified improvement rather than buy a platform to compensate for their absence.
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.
Read-out
Derived maturity mapping combining the five evaluation categories of Grimson and Pyke in S44 with data-governance rules from S22 and machine-learning readiness guidance from S27. Coordinates are derived structural positions on an illustrative 0 to 10 ordinal scale, not empirical factory measurements.
Unit: derived maturity index (0 to 10 scale). Status: derived. Sources: S44, S22, S27.
A greenfield garment factory should advance its digital architecture through explicit maturity gates rather than calendar milestones or procurement schedules. In GFT's operating design, a plant deploys automated analytics, manufacturing execution platforms, and machine-learning models only after proving three preconditions: clean source capture, active management routines, and verified corrective actions.[S22][S27][S44] Buying a complex platform before a line reconciles daily piece counts, classifies downtime accurately, or closes a root-cause investigation does not accelerate maturity. Instead, premature automation encodes unverified assumptions into software, hides measurement error behind screens, and diverts engineering attention from the floor. Governing progression through four evidence-based stages, from controlled manual capture to active cadence, verified improvement, and a data-driven floor, protects capital and ensures that every digital tool rests on an auditable foundation.[S27][S43][S44]
The temptation to skip foundational discipline is strong in contemporary apparel manufacturing. Executive surveys report that 75 percent of fashion leaders prioritise artificial intelligence at scale, encouraged by sector estimates showing inventory costs falling by 35 to 40 percent, forecast errors declining by 20 to 50 percent, and lost sales dropping by up to 65 percent.[R27] Those projections, estimating US$150 to 275 billion in sector operating-profit potential over three to five years, describe mature supply chains holding years of transaction history.[R27] They are not a forecast for a new cut, make, trim, and finishing plant. In Cameroon, where as of the 31 January 2024 programme cut-off about 95 percent of cotton fibre is exported raw against about 5 percent processed locally, and where policy targets 50 percent local processing by 2030, a greenfield plant starts without an inherited database.[R06][R07] Ramp-up research shows that a new process begins poorly understood, with low yields and production rates while teams learn the work.[S11]
13.1
Operational maturity is a joint property of human routines, organisational accountability, and technical infrastructure, not a count of software licences. When management evaluates readiness solely by inspecting information-technology assets, a factory can look advanced because it displays office charts while supervisors still dispute bundle counts on paper. Research on sales and operations planning maturity by Grimson and Pyke avoids this distortion by evaluating five evolutionary stages across five categories: meetings and collaboration, organisational structure, measures of performance, information technology, and integrated planning.[S44] A factory that installs five-stage planning software while running one-stage reactive meetings remains functionally immature because it cannot turn outputs into physical adjustments.[S44]
GFT adapts this five-domain structure to govern a greenfield factory across its first eighteen months, an illustrative horizon detailed in the closing roadmap.[S44] GFT binds all five domains to four sequential gates: Gate One (clean manual capture), Gate Two (active management cadence), Gate Three (verified improvement loop), and Gate Four (data-driven floor). Each gate requires proof across every domain before capital or engineering time is released for the next layer.[S22][S43][S44] Production, the warp of the factory, and measurement, the weft, remain under equal tension as the plant scales.
When a plant upgrades information technology faster than performance measures or meeting discipline, three faults appear. First, unvalidated source inputs travel faster through automated pipelines than through manual review, turning local recording mistakes into plant-wide scheduling errors.[S23][S24] Second, supervisors disengage from reviews when digital tools present composite indices or opaque recommendations that conflict with workstation constraints.[S07][S46] Third, commercial and finance teams treat software outputs as authoritative forecasts even when standard minute values or first-pass yield denominators remain provisional during style ramp-up.[S01][S11]
To make this interdependence inspectable, Exhibit 60 maps the five maturity domains from the sales and operations planning literature alongside GFT's four gates across process and governance discipline on the horizontal axis, and technical and analytical complexity on the vertical axis.[S22][S27][S44]
The derived positions in Exhibit 60 explain why GFT sequences governance ahead of technical complexity.[S22][S27][S44] The upper-left quadrant represents premature automation, where a factory invests in complex software while governance discipline stays below the midpoint. GFT's path moves first into the lower-right governed foundation quadrant during Gate One and Gate Two, establishing meeting discipline, role accountability, and metric definitions before crossing into the upper-right data-driven quadrant at Gate Three and Gate Four.[S27][S44] While Exhibit 60 clarifies structural order across domains, it cannot establish the exact calendar duration any specific factory will require to move between gates.
No single department holds unilateral authority to declare a gate passed. Data governance guidance defines governance as setting and enforcing priorities for managing and using data as a strategic asset, requiring explicit decision rights and accountable roles.[S22] GFT assigns gate review authority jointly to the plant director, the quality and industrial-engineering leads, the human-resources lead, and the data steward. If the information-technology lead shows that a dashboard ingests sewing counts every five minutes, an illustrative polling interval, but the quality lead shows that inspectors disagree on defect grading in 25 percent of repeat checks during an illustrative gauge study, the gate stays closed.[S24][S40] Performance standards must be measurable, understandable, verifiable, equitable, and achievable across all teams before the system scales.[S45]
13.4
A controlled manual or lightweight digital log is not an awkward stopgap; it is the factory's first genuine digital asset. Garment-floor studies show that front-line data in multi-variety, small-batch apparel lines can be captured through radio-frequency identification (RFID) systems and sewing-machine power monitors.[S19][S20] Yet raw event streams become useful for control charts or work-time extraction only when linked to defined operations, tasks, and batches.[S19][S20] Without an unambiguous counting boundary, an RFID scan merely records that a bundle tag passed an antenna; it does not prove that the operation met first-pass quality rules or followed the specified method.[S01][S03][S10]
GFT treats Gate One as a test of semantic and procedural control rather than hardware spend. During the first months of production, a structured workbook or form-entry tool is acceptable provided it enforces the data-quality model in ISO/IEC 25012 and the six core data-quality dimensions defined by the UK Government Data Quality Framework: completeness, validity, consistency, timeliness, uniqueness, and accuracy.[S23][S24] When a clerk or supervisor logs shift output, downtime minutes, defect codes, and meter reads in a governed table that checks every identifier against master data, the factory is building a machine-readable repository.[S21][S25]
For a manual or lightweight capture tool to qualify as a Gate One asset, it must follow five rules:
GFT note
GFT recommends keeping the Gate One capture schema identical to the table structure of a future manufacturing execution system. Using the same primary keys, effective-date columns, and lookup lists in day-one workbooks avoids data-cleansing projects later. When the plant migrates to a database or edge service, historical rows import directly.[S21][S25]
Once manual capture runs reliably, targeted sensing can cut clerical effort or sharpen time resolution. Attaching a power monitor to bottleneck sewing machines lets engineers extract stitching and idle times automatically, while bundle-level RFID readers log work-in-progress movements across cutting, sewing, and finishing.[S19][S20] In GFT's gate sequence, sensors are tested against the manual baseline rather than deployed alone. If a power trace reports 380 running minutes on a shift while the downtime log implies 420 running minutes, an illustrative 40-minute gap, that discrepancy triggers a check of threshold calibration and logging timeliness.[S20][S40] Because the metric contract exists, the sensor strengthens the system of record.[S01][S22]
13.7
Read-out · illustrative
Illustrative: GFT hour-by-hour line-board layout for one 8-hour sewing shift (40 operators, 24.0 SMV woven trouser, ramp cohort week 4 at 65 percent target efficiency). Values are illustrative design entries showing how balanced in-process and end-of-process measures (S02), daily line balancing (S04) and unmerged downtime causes (S07) appear with an explicit owner and response window. Not an observed GFT factory result.
Unit: illustrative shift-board control blocks. Status: illustrative. Sources: S02, S04, S07. Illustrative figures are teaching examples, not measured factory results.
Read this exhibit in chapter 11 ↗Clean data sitting in an unreviewed table creates storage cost, not control. Gate Two requires the factory to prove that its teams use captured records on a fixed cadence to make decisions, allocate resources, and close gaps. In the sales and operations planning maturity model, meetings and collaboration form a core domain alongside organisational structure and performance measures.[S44] Moving from reactive firefighting to coordinated management depends on whether teams meet with shared facts, test assumptions, and leave with owned actions.[S42][S44]
Active management spans three nested horizons that must work before predictive software is considered. Daily, line supervisors hold shift-start huddles using balanced in-process and end-of-process measures to review prior-shift good output, first-pass yield, downtime causes, and operator skill coverage, rebalancing stations during the shift as flow changes.[S02][S04][S05] Weekly, operations, quality, and maintenance leaders review recurring losses and cross-line constraints, applying a systematic improvement method rather than reading composite overall equipment effectiveness (OEE) scores alone.[S06][S07][S08] Monthly, senior leadership conducts sales and operations planning at the product-family level, reconciling delivery promises with demonstrated line capability, material supply, and workforce ramp curves.[S11][S42]
Confusing visual reporting with active management is a common digitalisation failure. A sponsor may install wall screens showing colour-coded hourly targets, yet find months later that the same sewing machines fail repeatedly for the same thread-break or feed-dog faults. OEE literature warns that measuring availability, performance, and quality without a systematic improvement framework fails to produce lasting gains, while merged downtime categories obscure specific losses.[S07][S08] Balanced-scorecard research likewise notes that a rigid top-down cascade can fail in adaptive settings when supervisors cannot challenge flawed standards or escalate bottlenecks.[S46]
Gate Two audits behavioural execution across four operating trails:
13.9
1. Define decision and guardrail
1. Define decision and guardrail. Starting point. Leads to 2. Name accountable user (Actionable lever). · illustrative
Machine-learning engineering guidance states a primary rule: if machine learning is not required for a product or decision, do not use it until data exists.
2. Name accountable user
2. Name accountable user. From 1. Define decision and guardrail (Actionable lever). Leads to 3. Audit data completeness (Cadence assigned). · illustrative
3. Audit data completeness
3. Audit data completeness. From 2. Name accountable user (Cadence assigned). Leads to 4. Test simple heuristic first (Contract met). · illustrative
4. Test simple heuristic first
4. Test simple heuristic first. From 3. Audit data completeness (Contract met). Leads to 5. Verify out-of-sample lift (Gap remains). · illustrative
5. Verify out-of-sample lift
5. Verify out-of-sample lift. From 4. Test simple heuristic first (Gap remains). Leads to 6. Stage with rollback switch (Beats heuristic). · illustrative
6. Stage with rollback switch
6. Stage with rollback switch. From 5. Verify out-of-sample lift (Beats heuristic). Leads to 7. Monitor drift and benefit (Live shadow test). · illustrative
7. Monitor drift and benefit
7. Monitor drift and benefit. From 6. Stage with rollback switch (Live shadow test). End point. · illustrative
The method behind it: Gate readiness and heuristic-comparison check.
Illustrative GFT decision sequence for qualifying candidate analytical and machine-learning use cases. Design logic synthesises the data-first rule in S27, phased MLOps automation in S28, and lifecycle trustworthiness evaluation in S1301.
Unit: sequential qualification gates. Status: illustrative. Sources: S27, S28, S1301. Illustrative figures are teaching examples, not measured factory results.
13.9 · the bodyExhibit 61 above; its figures are illustrative where marked.
Machine-learning engineering guidance states a primary rule: if machine learning is not required for a product or decision, do not use it until data exists.[S27] In a greenfield garment factory, almost every core decision during the first year, including line balancing, bundle release, defect containment, preventive maintenance scheduling, and submeter reconciliation, can be handled through deterministic arithmetic, statistical process control, and structured review.[S04][S09][S27] History-dependent models need representative, accurately labelled training distributions. During early ramp-up, the process shifts week by week as operators build skill, engineers refine methods, and fabric batches vary.[S03][S11] Training a predictive model on unstable ramp-up logs yields brittle weights that fail once the line nears steady state.
GFT enforces Gate Three and Gate Four use-case qualification so model count never becomes a vanity target. Even when market reports project rapid growth in specialised artificial-intelligence segments, such as the fabric-defect-detection AI market estimated at US$523.4 million in 2024 and projected to reach US$6,522.1 million by 2033 at 33.8 percent compound annual growth, a factory must separate pre-trained optical sensing from plant-specific historical forecasting.[R28] Inspection studies report that manual fabric inspection is 60 to 70 percent accurate, missing 20 to 30 percent of defects, whereas line-scan computer vision running at up to 60 metres a minute can exceed 90 percent accuracy across more than 40 defect types.[R29] Yet even a pre-trained camera needs local calibration against the plant's weaves, dye shades, defect taxonomy, and human adjudication logs before its outputs govern acceptance.[R29][S25][S40]
Before any machine-learning or analytical use case receives budget, GFT subjects the proposal to a seven-step gate sequence combining machine-learning rules, gradual MLOps controls, and the trustworthiness functions (govern, map, measure, manage) of the NIST AI Risk Management Framework.[S27][S28][S1301] Exhibit 61 shows this qualification flow.
As shown in Exhibit 61, the sequence filters out weak proposals early.[S27][S28][S1301] At Step 1 and Step 2, the proposer identifies the decision the model will change, the guardrail metrics, and the role owner who will act on the output within a daily, weekly, or monthly forum.[S02][S22][S44] At Step 3, the data steward audits whether historical records meet completeness, validity, consistency, uniqueness, and measurement-error thresholds across comparable cohorts.[S24][S40] Step 4 requires building and testing a transparent heuristic or statistical baseline first, such as a rolling cohort average, a standard minute bottleneck check, or a control chart.[S03][S09][S27] Only when Step 5 proves meaningful lift over that heuristic on held-out chronological cohorts does the model advance to staged shadow deployment (Step 6) and drift and benefit monitoring (Step 7).[S28][S43][S1301] Because Exhibit 61 is an illustrative workflow, it defines procedural checkpoints, not a guarantee of financial return without local testing.
Interlude · the cloth half-woven
Woven: sections 13.0 to 13.9Still to weave: 13.11 to 13.15
Read-out
Illustrative worked example. 64 OTIF orders + 16 orders not OTIF = 80 orders due. OTIF rate = 64 / 80 x 100 = 80 percent. These are assumed values used to demonstrate the calculation.
Unit: illustrative orders and percent. Status: illustrative. Illustrative figures are teaching examples, not measured factory results.
Read this exhibit in its chapter ↗13.11
Layer 4: Conditional pipeline automation
Layer 4: Conditional pipeline automation: Scheduled retraining triggers, Automated feature pipelines, Continuous model delivery · derived
When a candidate use case clears every qualification gate, such as an optical fabric-defect classifier or a family-level demand model once multi-season history exists, the factory needs controls to keep the model reliable in production.
Layer 3: Production monitoring and rollback
Layer 3: Production monitoring and rollback: Outcome and drift tracking, Human adjudication sampling, One-step heuristic fallback · derived
Layer 2: Model governance and evaluation
Layer 2: Model governance and evaluation: Named model and decision owner, Chronological cohort validation, Heuristic baseline benchmark · derived
Layer 1: Data and artefact foundation
Layer 1: Data and artefact foundation: Immutable dataset snapshots, Schema and range validation, Code and parameter versioning · derived
Derived four-layer control architecture combining gradual MLOps implementation principles from S28, structured data-quality requirements from S23, and AI risk-management evaluation functions from S1301. Layers 1 to 3 are mandatory at first model deployment; Layer 4 is conditional on portfolio scale.
Unit: staged control layers. Status: derived. Sources: S28, S23, S1301.
13.11 · the bodyExhibit 62 above; its figures are illustrative where marked.
When a candidate use case clears every qualification gate, such as an optical fabric-defect classifier or a family-level demand model once multi-season history exists, the factory needs controls to keep the model reliable in production.[R27][R29] Machine-learning operations (MLOps) should not start with a heavy orchestration platform. Cloud guidance notes that teams can implement MLOps practices gradually to improve automation of model development and production.[S28] For a site with a small technical team, starting with automated continuous-retraining pipelines adds fragility and can let a model retrain on corrupted sensor bursts or unlabelled style changes.[S24][S28]
GFT sequences MLOps maturity through staged controls aligned with the NIST AI Risk Management Framework functions of governing accountability, mapping context, measuring performance and drift, and managing incidents through fallback routines.[S28][S1301] Initially, the plant uses script-backed manual MLOps controls focused on reproducibility and safety. Automated retraining and feature pipelines follow only when model count and retraining frequency make manual triggers a constraint.[S28]
To separate mandatory day-one governance from optional later automation, Exhibit 62 arranges MLOps responsibilities into a four-layer architecture.[S23][S28][S1301] The bottom three layers are compulsory before any model influences a factory decision; the top layer is conditional on scale.
Reading Exhibit 62 from bottom to top shows how engineering controls prevent silent model decay:[S23][S28][S1301]
While Exhibit 62 defines the hierarchy of MLOps controls, it cannot specify the exact retraining interval for a given factory model, which depends on empirical drift across seasonal fabric and style changes.
13.13
A digital architecture that assumes unbroken high-bandwidth internet links, unlimited cloud budgets, and resident vendor engineers will fail under the conditions of a greenfield site. Operational technology (OT) security and reliability guidance from NIST emphasises that industrial systems face distinct performance, reliability, and safety requirements compared with office systems.[S1302] On a garment floor, bundle ticket printing, cutting-table marker release, in-line quality logging, and utility submeter accumulation cannot halt because an external telecom link lags or a cloud service enters maintenance.
To match architecture to site constraints, GFT applies the distributed computing principles of the NIST Fog Computing Conceptual Model.[S26] Fog and edge computing place applications, management, and analytics inside the local network using a distributed model, shortening the path between capture and processing.[S26] Running local validation, buffering, and time-critical inference on site-resident nodes keeps shift operations running during wide-area outages, while synchronising summaries to central storage when connectivity returns.[S26][S1302]
When sponsors decide whether to build an internal tool, configure open-source software, or buy a commercial suite, procurement reviews too often focus on vendor demonstrations. Public-sector technology guidance notes that choosing technology shapes a team's ability to operate and iterate its service, and advises using open standards to manage interfaces between components so later changes remain practical.[S1303] Similarly, NIST smart-manufacturing standards research stresses that exchanging and using product, production, and business data across disparate systems depends on information standards.[S21]
GFT applies four engineering tests to every build-against-buy choice:
Exhibit 63 applies these criteria across four functional subsystems of a greenfield garment factory.
Read-out
| Source capture and line boards | Governed workbook or local web form | Modular edge capture with RFID or power inputs | Full shift operation on local LAN with buffered sync | Daily automated export of raw events in open schema |
|---|---|---|---|---|
| Shop-floor MES and routing | Master-data tables plus paper bundle travellers | Configurable MES module bound to metric contracts | Local execution of work orders and bundle tracking | Complete routing and WIP history exportable without vendor scripts |
| Fabric and garment vision | Standardised manual grading and Gauge R&R checks | Edge camera inference writing to shared defect taxonomy | Local on-device inference at line speed without cloud lag | Retained image samples and labels stored in plant repository |
| Demand and S&OP analytics | Deterministic capacity and SMV reconciliation sheets | Statistical or ML forecast benchmarked on heuristic | Monthly batch run executable on local workstation | Model weights, features and SQL logic owned by plant |
Illustrative GFT option assessment for four factory subsystems, applying smart-manufacturing interoperability standards from S21, data-governance ownership rules from S22, local fog-computing resilience from S26, and open-standards interface guidance from S1303.
Unit: architectural evaluation matrix. Status: illustrative. Sources: S21, S22, S26, S1303. Illustrative figures are teaching examples, not measured factory results.
The progression across columns in Exhibit 63 shows why architecture should grow with maturity rather than begin at maximum vendor lock-in.[S21][S26][S1303] For source capture and shop-floor execution, Gate One and Gate Two rely on governed tables and local forms that cost little and let the factory refine operation codes and downtime taxonomies through shift experience.[S07][S25] At Gate Three and Gate Four, the plant can extend modular local services or license a commercial MES module, knowing which schemas and offline functions the vendor must support.[S21][S26] For computer vision, where line-scan inspection runs at up to 60 metres a minute, inference executes on local edge hardware beside the frame because wide-area network latency cannot match moving fabric, while labelled images stay in the plant's repository so future models are not tied to one vendor.[R29][S26][S1303] Because Exhibit 63 is an illustrative comparison, it is not procurement, legal, or contract advice; sponsors must assess local telecom reliability, power quality, and internal engineering skills before signing contracts.
13.15
Related · chapter 116 paragraphs step it
Read-out
Illustrative: worked arithmetic bridge for a four-line garment family (160 operators, 22 days, 480 minutes per shift = 1,689.6 thousand gross minutes), deducting planned huddles/drills/absence (-84.5), unplanned downtime (-160.5), ramp speed loss (-216.7) and rework loss (-85.9) to reach 1,142.0 thousand net deliverable standard minutes (47,583 garments at 24.0 SMV) against a 1,320.0 thousand standard-minute demand request. Demonstrates monthly S&OP reconciliation (S42) using SMV (S03), OEE components (S06) and ramp constraints (S11). Not an observed GFT factory result.
Unit: thousand standard minutes per month. Status: illustrative. Sources: S03, S06, S11, S42. Illustrative figures are teaching examples, not measured factory results.
Read this exhibit in its chapter ↗The final maturity rule protects the factory when it upgrades its software stack: moving from a Gate One workbook to a Gate Three database or a Gate Four manufacturing execution system is a storage transition, not a reset of operational meaning. In poorly governed projects, a vendor arrives with pre-built KPI definitions that quietly change how availability, first-pass yield, standard minutes, or water consumption are calculated. If the new platform counts repaired garments as first-pass output, merges waiting time into breakdown time, or drops the distinction between water withdrawal and water consumption, the factory loses its historical baseline.[S06][S07][S10][S34] Management can no longer tell whether a post-migration rise in efficiency reflects physical improvement on the lines or a silent shift in software arithmetic.[S05][S43]
GFT prevents this loss of institutional memory by treating the data governance framework, ISO 22400 metric contracts, and W3C-aligned master-data vocabularies as the constitution of the factory's performance system.[S01][S22][S25] Any incoming software platform, whether built in-house or licensed, must conform to the factory's existing definitions rather than overwrite them.
Before an upgraded database, MES, or analytics service becomes the system of record, the data steward and business owners run a four-step migration protocol:
Binding technical ambition to behavioural and evidentiary proof turns digital strategy from a speculative purchase into an engineering discipline. A greenfield garment factory does not need artificial intelligence on its first shift; it needs clear counting rules, honest downtime and defect logs, and supervisors who use daily evidence to balance lines and support workers.[S01][S04][S27] Once that warp and weft are tight, each subsequent layer, from edge buffering and machine sensing to statistical forecasting and computer vision, fits into a governed structure that proves whether the technology improved cost per good garment, delivery reliability, or resource intensity.[S19][S20][S26][S43] The next chapter translates these four gates into an eighteen-month roadmap of deliverables, benefit registers, and board scorecards. This guidance is an engineering framework and does not constitute investment, legal, or tax advice.
Check the record
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.
40 sources · 258 citation marks on this pageMost cited: S27, 16 times
Grade AGrade BGrade CKnot size: citations on this page
All 40 records shown
R06 · Grade C · inherited record
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.
R07 · Grade C · inherited record
Inherited evidence R07
SND30 restates the 50 percent local-processing target by 2030.
Publication date unrecorded. Inherited source-text record. The original publication has not been independently reopened for this web edition.
R27 · Grade B · inherited record
Inherited evidence R27
AI at scale in fashion: inventory cost down 35 to 40 percent; forecast accuracy up 20 to 30 percent; revenue per customer up 15 to 25 percent; forecast errors cut 20 to 50 percent; lost sales down up to 65 percent; 75 percent of executives prioritising; sector operating-profit potential US$150 to 275 billion over three to five years.
Publication date unrecorded. Inherited source-text record. The original publication has not been independently reopened for this web edition.
R28 · Grade C · inherited record
Inherited evidence R28
Fabric-defect-detection AI market US$523.4 million (2024) to US$6,522.1 million (2033) at 33.8 percent a year; Asia-Pacific about 46 percent in 2024.
Publication date unrecorded. Inherited source-text record. The original publication has not been independently reopened for this web edition.
R29 · Grade C · inherited record
Inherited evidence R29
Manual defect inspection 60 to 70 percent accurate, missing 20 to 30 percent of defects; computer vision above 90 percent across more than 40 defect types; line-scan inspection up to 60 metres a minute.
Publication date unrecorded. Inherited source-text record. The original publication has not been independently reopened for this web edition.
c13-S01 · Grade A
ISO 22400-2:2014 Automation systems and integration: Key performance indicators (KPIs) for manufacturing operations management, Part 2: Definitions and descriptions
The KPIs are presented by means of their formula and corresponding elements, their time behaviour, their unit/dimension and other characteristics.
2014. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S02 · Grade B
Lean Manufacturing Techniques for Ready Made Garments Industry
The best approach is to balance the metrics you use among the three categories (financial, behavioral, and core-process) and to use a mix of in-process and end-of-process metrics.
2017. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S03 · Grade B
Sector systems analysis of textiles and clothing subsector in Ghana
A particularly important calculation is the Standard Minute Value (SMV), which is defined as “the time value arrived at for a task based on the average rate of output which qualified workers will naturally achieve without over exertion provided that they know and adhere to the specified method and provided that they are motivated to apply themselves to their work.“
2023. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S04 · Grade B
Implementation of the Factory Improvement Toolset in the Asian garment sector: Final report
Line balancing is done by line supervisors systematically, before production through planning, and during production, daily.
2023. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S05 · Grade B
Advancing productivity, quality improvement and decent work in Ethiopia's garment industry - results so far.
As step zero, SCORE trainers conduct a baseline assessment to identify existing gaps and opportunities for improvement in SMEs, and measure change through Key Performance Indicators (KPIs).
2022-06-01. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S06 · Grade B
Comparison of OEE-Based Manufacturing Productivity Metrics
The OEE measurement aims to detect production losses and maximise equipment effectiveness.
2023. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S07 · Grade B
Uncovering the Influence of Operational Factors on Manufacturing Efficiency with Real Time Data
However, this loss categorization merges all downtime events, which can obstruct the identification of specific losses, representing a significant limitation of OEE.
2025. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S08 · Grade B
Overall equipment effectiveness: a review and development of an integrated improvement framework
The review found that there is a lack of improvement frameworks with systematic steps specifically developed for OEE implementation.
2020-05-11. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S09 · Grade A
e-Handbook of Statistical Methods: What are Process Control Techniques?
The underlying concept of statistical process control is based on a comparison of what is happening today with what happened previously.
2003. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S10 · Grade A
First Pass Yield Improvement Creates Dramatic Reduction in Lead Time
In addition, total processing time went from 11.8 hours to 3.4 hours and first pass yield jumped from 10% to 90%.
2022-08-04. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S11 · Grade B
Learning and process improvement during production ramp-up
During that time, the new production process is ill understood, which causes low yields and low production rates.
2001. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S12 · Grade B
Evaluation of Manual Skill Degradation Due to Automation in Apparel Manufacturing
Therefore, capacity development drills of sewing operators are essential to maintain an efficient required skill set.
2021-11-23. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S17 · Grade B
Piece rate pay brings mixed results for garment workers, new study shows
The research found that workers whose pay is jointly determined by hourly pay and a piece rate (described as partial piece rate pay) are more likely to report reduced emotional and physical health compared to workers paid by the hour.
2018-12-20. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S18 · Grade B
Excessive Overtime, Workers and Productivity: Evidence and Implications for Better Work
It has been shown that excessive overtime can lead to poorer worker health, reduced efficiency, high rates of accidents, and worker dissatisfaction, resulting in high worker turnover.
2011. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S19 · Grade B
A method of garment factory workers' performance monitoring using control chart based on RFID system
Massive production data of front-line workers are recorded every day by RFID system in garment factory, which are mainly in multiple variety and small batch production.
2019-09-03. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S20 · Grade B
Remote Sensing of Sewing Work Levels Using a Power Monitoring System
In this study, a power monitoring system attached to the sewing machine was used to remotely collect power consumption data, which were then analyzed to extract the working times for a series of sewing tasks.
2020-04-29. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S21 · Grade A
NIST IR 8107: Current Standards for Smart Manufacturing Systems
The ability of disparate systems, however, to exchange, understand, and exploit product, production, and business data rests critically on information standards.
2016. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S22 · Grade A
Data Governance Playbook
Data governance is the process of setting and enforcing priorities for managing and using data as a strategic asset.
2020. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S23 · Grade A
ISO/IEC 25012:2008 Software engineering: Software product Quality Requirements and Evaluation (SQuaRE): Data quality model
This International Standard defines a general data quality model for data retained in a structured format within a computer system.
2008. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S24 · Grade A
The Government Data Quality Framework
There are six core data quality dimensions, as defined by DAMA UK.
2020-12-03. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S25 · Grade A
Data on the Web Best Practices
Finally, to promote the interoperability among datasets it is important to adopt data vocabularies and standards.
2017-01-31. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S26 · Grade A
NIST SP 500-325: Fog Computing Conceptual Model
One solution is to decentralize applications, management, and data analytics into the network itself using a distributed and federated compute model.
2018-03-14. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S27 · Grade B
Rules of Machine Learning: Best Practices for ML Engineering
If machine learning is not absolutely required for your product, don't use it until you have data.
2025-08-25. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S28 · Grade B
MLOps: Continuous delivery and automation pipelines in machine learning
You can gradually implement these practices to help improve the automation of your ML system development and production.
2024-08-28. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S34 · Grade A
GRI 303: Water and Effluents 2018
Water consumption measures water used by an organization such that it is no longer available for use by the ecosystem or local community in the reporting period.
2018. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S40 · Grade A
e-Handbook of Statistical Methods: Gauge R & R studies
The purpose of this section is to outline the steps that can be taken to characterize the performance of gauges and instruments used in a production setting in terms of errors that affect the measurements.
2003. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S41 · Grade B
How to Test Your A3 Thinking
Reading an A3 in reverse helps A3 owners and their leaders confirm the use of a logical, fact-based thinking process.
2022-08-10. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S42 · Grade B
A case method for Sales and Operations Planning: a learning experience from Germany
It is performed at least once a month and is reviewed by management at an aggregate (product family) level.
2017. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S43 · Grade A
Local Authority Major Schemes: Benefits Management and Evaluation Framework
It also ensures those benefits are managed and monitored beyond the point of project completion.
2025. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S44 · Grade B
Proposal of a model for sales and operations planning (S&OP) maturity evaluation
The framework elaborated by Grimson & Pyke (2007), consists of five evolutionary stages, analyzed in the following categories: meetings and collaboration, organizational structure, measures of performance, information technology, and integrated planning.
2017-11-27. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S45 · Grade A
Performance Management Cycle
Performance elements and standards should be measurable, understandable, verifiable, equitable, and achievable.
2019-11-04. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S46 · Grade B
Balanced Scorecard: History, Implementation, and Impact
Another criticism focuses on the BSC’s presumption of a rational, top-down approach to strategy execution, which may not align with decentralized or adaptive organizational structures.
2025-03-19. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S1301 · Grade A
AI Risk Management Framework
The NIST AI Risk Management Framework (AI RMF) is intended for voluntary use and to improve the ability to incorporate trustworthiness considerations into the design, development, use, and evaluation of AI products, services, and systems.
2023-01-26. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S1302 · Grade A
SP 800-82 Rev. 3, Guide to Operational Technology (OT) Security
This document provides guidance on how to secure operational technology (OT) while addressing their unique performance, reliability, and safety requirements.
2023-09-28. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c13-S1303 · Grade A
Choosing technology: an introduction
If they’re available you should use open standards to manage the interfaces between parts of your system. Using open standards makes it easier to make changes later.
2024-10-15. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗S22 United States Federal Data Strategy. 2020.
S27 Google for Developers. 2025-08-25.
S44 Production, SciELO. 2017-11-27.
S43 UK Department for Transport. 2025.
R27 McKinsey, The State of Fashion 2025. Publication date unrecorded; inherited source-text check.
R06 Cameroon Trade Hub. Publication date unrecorded; inherited source-text check.
R07 Ecofin Agency. Publication date unrecorded; inherited source-text check.
S11 International Journal of Production Economics, Elsevier. 2001.
S23 ISO/IEC. 2008.
S24 UK Government Data Quality Hub. 2020-12-03.
S07 Scientific Reports, Springer Nature. 2025.
S46 Encyclopedia, MDPI. 2025-03-19.
S01 International Organization for Standardization. 2014.
S40 NIST/SEMATECH. 2003.
S45 United States Office of Personnel Management. 2019-11-04.
S19 The International Journal of Advanced Manufacturing Technology, Springer Nature. 2019-09-03.
S20 Applied Sciences, MDPI. 2020-04-29.
S03 International Labour Organization. 2023.
S10 National Institute of Standards and Technology, Manufacturing Extension Partnership. 2022-08-04.
S21 National Institute of Standards and Technology. 2016.
S25 W3C. 2017-01-31.
S06 Diamond Congress, Creative Construction Conference. 2023.
S09 NIST/SEMATECH. 2003.
S42 Production, SciELO. 2017.
S02 International Labour Organization. 2017.
S04 International Labour Organization. 2023.
S05 International Labour Organization. 2022-06-01.
S08 International Journal of Productivity and Quality Management, Inderscience. 2020-05-11.
S41 Lean Enterprise Institute. 2022-08-10.
R28 DataIntelo. Publication date unrecorded; inherited source-text check.
R29 MDPI; industry inspection literature. Publication date unrecorded; inherited source-text check.
S28 Google Cloud. 2024-08-28.
S1301 National Institute of Standards and Technology. 2023-01-26.
S17 ILO and IFC, Better Work. 2018-12-20.
S18 ILO and IFC, Better Work. 2011.
S12 Applied Sciences, MDPI. 2021-11-23.
S1302 National Institute of Standards and Technology. 2023-09-28.
S26 National Institute of Standards and Technology. 2018-03-14.
S1303 UK Government Digital Service. 2024-10-15.
S34 Global Reporting Initiative. 2018.
Keep the thread · chapter 13 of 14
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
01Part I: The measurement contract
A factory without an operating baseline cannot rely on hindsight, so GFT treats measurement as part of production from the outset.
02Part I: The measurement contract
Early targets should describe a controlled learning trajectory, with yield, rate and capability observed together rather than assumed away.
03Part I: The measurement contract
Strategic intent becomes an operating system when each line measure carries a decision, an owner, a feasible record and a guardrail.
04Part I: The measurement contract
Operational definitions, data-quality controls and measurement-system analysis must precede target comparisons and automated analytics.
05Part II: Measure the operating system
Good output is the result, while the production system must retain the component losses that explain why the result occurred.
06Part II: Measure the operating system
The quality system should connect first-pass yield, defect causes and variation so that inspection leads to prevention rather than sorting alone.
07Part II: Measure the operating system
Standard work, balanced capacity and order-level flow records must support the delivery promise before expediting becomes routine.
08Part II: Measure the operating system
A low nominal wage does not establish a low cost per good garment, so the people scorecard must include skill, supervision, attendance and working-time effects.
09Part II: Measure the operating system
Energy, water, waste and emissions become manageable only when each quantity has a declared boundary, source record and production denominator.
10Part III: Turn records into decisions
Source capture, master data and explicit stewardship turn daily logs into an operational asset without requiring a heavy platform on day one.
11Part III: Turn records into decisions
Daily correction, weekly problem solving and monthly reconciliation give each signal a forum, an authority level and a recorded response.
12Part III: Turn records into decisions
A corrective-action system should distinguish abnormal variation, test causes, verify effects and update standard work rather than count completed actions.
13Part IV: Scale with control
The factory should earn each digital step through reliable records, active management and verified improvement rather than buy a platform to compensate for their absence.
14Part IV: Scale with control
GFT recommends four gated stages: clean capture, active cadence, verified improvement and a data-driven floor, with value measured in the same system throughout.