Research libraryThe Industrial Data Spine
Chapter 05 of 14 · Part II · Draft web edition
A durable architecture preserves source evidence, adds governed context once and lets each decision consume a controlled product.
The cloth is woven from this chapter's 5 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 minimum architecture separating capture, immutable raw evidence, contextual canonical records, published data products and operational decisions, based on interface boundaries in S05, provenance modelling in S09 and S0501, smart manufacturing standards in S10 and pipeline lineage in S15.
Unit: . Status: derived. Sources: S05, S09, S10, S15, S0501.
Overture · 1 min
The smallest durable industrial data spine is a three-stage separation of responsibilities, not a large software suite. It separates capture close to physical work, context assembly in a canonical operational core, and consumption through read-only, purpose-specific data products.[S05][S10] In Cameroon, where the inherited sector baseline prepared on 31 January 2024 records about 95 percent of cotton fibre exported raw and about 5 percent processed locally against a 50 percent target by 2030 (original dates unrecorded and not refreshed in 2026), domestic mills cannot afford brittle data estates.[R06][R07] Reported capacity utilisation at domestic textile plants in July 2023 ranged from 10.83 percent at CICAM 2 to 33.7 percent at Garoua and 48.9 percent at CICAM 1, with underutilisation costing about CFAF 2 billion in 2020.[R12] Connecting shop-floor logs, machine counters, quality sheets and order schedules directly to spreadsheets or dashboards without separating raw evidence from business transformation means every formula change alters history and every new consumer multiplies integration effort.[S18][S39] Separating capture, context and consumption lets a lean team preserve immutable shop-floor evidence, attach governed meaning once, and serve management reporting, traceability and machine-learning pipelines from one auditable foundation.[S08][S09][S15]
5.1
Related · chapter 045 paragraphs step it
Read-out
Illustrative scoring model (0 = untested, 1 = manual spot check, 2 = entry syntax check, 3 = automated gate with quarantine, 4 = versioned gate with full lineage) across four factory record sets and five quality dimensions, derived from ISO/IEC 25012 (S03), ISO/IEC 5259-3 (S04) and W3C DQV (S0401). Values are illustrative and not a measured GFT factory result.
Unit: illustrative coverage score (0 to 4). Status: illustrative. Sources: S03, S04, S0401. Illustrative figures are teaching examples, not measured factory results.
Read this exhibit in its chapter ↗Capture must occur at the physical boundary where materials move, machines cycle and operators inspect garments, and it must continue working when external networks fail.[S22][S33] An event loses diagnostic fidelity when an operator must remember a stoppage until shift end or transcribe a chalkboard tally into a terminal hours later.[S19] Keeping capture adjacent to work means placing manual entry forms at the sewing line or inspection table, reading programmable logic controllers and utility meters at the cell boundary, and running edge vision inference next to the fabric frame rather than across an unstable wide-area link.[R29][S05][S22] Because operational technology environments have strict performance, reliability and safety constraints, capture components must observe production passively or through defined interfaces without delaying control loops or introducing unsafe dependencies.[S05][S22]
Recording an event where it happens requires local resilience against network interruptions.[S33] Where telecommunications or switch faults occur intermittently, a capture node that depends on continuous cloud or central-server availability drops production evidence when operational stress is highest.[S22][S33] Product-specific edge architectures illustrate the required mechanism: when upstream connectivity is lost, a local parent device buffers device-to-cloud messages on durable storage until the connection returns.[S33] In GFT's minimum spine, that store-and-forward principle applies across all three capture channels: a line-side tablet caches signed shift entries locally, an edge gateway writes polled machine registers to a local ring buffer, and an optical inspection unit stores defect events and image hashes on local solid-state storage before upstream synchronisation.[R29][S08][S33]
Adjacency alone does not make captured records trustworthy unless the spine records how each value entered the system.[S09][S0501] Under the W3C PROV Data Model, provenance is information about entities, activities and agents involved in producing a piece of data, used to assess its quality, reliability and trustworthiness.[S0501] If a table stores a fabric defect count or stoppage duration without recording the generating agent and activity, downstream analysts and training pipelines cannot tell whether the figure came from a calibrated optical sensor, a manual tally sheet, a default fallback value or a model's inference.[S04][S09][S19]
Every captured record therefore carries an explicit capture-method and source-agent header alongside its payload.[S08][S09] Manual entries record the workstation role, entry mode and submission timestamp while respecting worker data minimisation by avoiding unnecessary personal identifiers.[S02][S29] Sensor and meter readings record the device identifier, interface protocol and polling interval.[S05][S06] Edge model outputs, such as automated fabric defect classifications where literature reports computer vision exceeding 90 percent accuracy across more than 40 defect types at line-scan speeds up to 60 metres a minute compared with 60 to 70 percent for manual inspection, must record the model identifier, version, confidence score and threshold applied.[R29][S15] Keeping manual, sensor and model-generated values structurally distinct at capture prevents a later pipeline from training a new model on the unverified predictions of an older model.[S04][S15][S18]
5.4
Once an event leaves the capture boundary, the spine must separate the duty to remember what was observed from the duty to present clean, contextualised data for daily decisions.[S09][S15] In many small factories, one database or shared workbook serves both roles: scripts or operators write incoming numbers into a table, and supervisors edit that same table in place when they spot a typo, a miscalibrated meter or a split order.[S18][S19] In-place mutation destroys the audit trail. When a downstream key performance indicator (KPI) shifts unexpectedly or a forecasting model degrades, the team cannot reconstruct what the capture layer reported on the day of production or test whether a cleaning rule introduced bias.[S04][S15][S19]
The minimum durable spine enforces a strict boundary between the immutable raw evidence log and the curated contextual store, as shown in Exhibit 20.[S05][S09][S15] The raw evidence layer accepts every receivable payload from capture, assigns an ingestion timestamp and cryptographic content hash, and writes the record to append-only storage.[S08][S09][S26] No user, steward or transformation script may update or delete rows in the raw layer during normal operations.[S25] Even when a record fails schema validation or violates a physical range check, the raw payload is preserved so engineers can inspect the exact malformed message, diagnose whether a controller register shifted or a manual form broke, and replay the raw stream once the parser or rule is fixed.[S01][S04][S15]
As Exhibit 20 illustrates, the five functional layers embody three architectural separations: capture is isolated from storage by local buffering and provenance headers; raw evidence is isolated from curated context by append-only immutability and quality gates; and curated context is isolated from consumption by read-only published products.[S05][S09][S10][S15] What Exhibit 20 cannot establish on its own is the physical deployment topology, such as whether layers two, three and four run inside a single relational database engine with schema-level access controls or across separate server nodes; that physical sizing belongs to the infrastructure design in WP18 rather than the governance architecture.[S05][S25]
Curated records support rapid, reliable consumption without forcing every dashboard query or feature pipeline to re-parse raw payloads and re-apply historical correction logs from scratch.[S04][S15][S0502] In the contextual layer, raw events are validated against their data contracts, linked to the master-data version active at the event time, and tagged with an explicit quality state such as observed, estimated, missing or manually corrected.[S03][S04][S08]
When a domain steward discovers that an operator entered the wrong lot code or that a power meter reported a spike during maintenance testing, the steward never overwrites the raw record and never silently edits the curated row in place.[S02][S09] Instead, the spine records a superseding correction event that references the original event identifier, states the corrected value and reason code, identifies the approving steward role, and links the new curated state to both the original raw entity and the correction activity using PROV-O derivation relationships.[S02][S09][S0501] Curated views expose the current effective state for routine reporting while retaining the full derivation chain so any historical report or model training snapshot can be reproduced as of the timestamp it was originally generated.[S09][S15][S0503]
5.7
Stage 1 (4x4 endpoints)
Stage 1 (4x4 endpoints): Point-to-point (S x C) 16, Canonical core (S + C) 8. Difference -8 (number of interface mapping contracts (illustrative)) · illustrative
Connecting each source directly to each consuming application creates an unmanageable web of bilateral translations.
Stage 2 (6x6 endpoints)
Stage 2 (6x6 endpoints): Point-to-point (S x C) 36, Canonical core (S + C) 12. Difference -24 (number of interface mapping contracts (illustrative)) · illustrative
Stage 3 (8x8 endpoints)
Stage 3 (8x8 endpoints): Point-to-point (S x C) 64, Canonical core (S + C) 16. Difference -48 (number of interface mapping contracts (illustrative)) · illustrative
Stage 4 (10x10 endpoints)
Stage 4 (10x10 endpoints): Point-to-point (S x C) 100, Canonical core (S + C) 20. Difference -80 (number of interface mapping contracts (illustrative)) · illustrative
The method behind it: Method.
Illustrative arithmetic comparing S + C canonical mappings with S * C bilateral point-to-point mappings across four symmetrical factory integration stages (4, 6, 8 and 10 sources and consumers). Structural rationale derived from S05, S10 and S39; counts are mathematical identities on assumed endpoint counts, not empirical measurements from a specific plant.
Unit: number of interface mapping contracts (illustrative). Status: illustrative. Sources: S05, S10, S39. Illustrative figures are teaching examples, not measured factory results.
5.7 · the bodyExhibit 21 above; its figures are illustrative where marked.
Connecting each source directly to each consuming application creates an unmanageable web of bilateral translations.[S10][S18] When a sewing plant has line-side logs, an order table, a quality sheet, a maintenance log and an energy meter, and feeds an overall equipment effectiveness (OEE) tracker, a buyer portal, a carbon calculator and a demand forecaster through point-to-point scripts, each script embeds private assumptions about shift boundaries, units, scrap codes and order identifiers.[S05][S38][S39] NIST's smart manufacturing analysis notes that standards are fundamental to delivering the right information at the right time, enabling action, and reducing adoption risk.[S10] Manufacturers in a 2025 NIST semiconductor workshop reported that data processing for a single application can reach 70 percent of total application cost when representations are not shared.[S39]
Point-to-point translation fails small teams because maintenance effort scales with the product of sources and consumers rather than their sum.[S10][S18] When an upstream machine tag or spreadsheet column changes, the custodian must update multiple extraction scripts, each written at a different time with incompatible filter rules.[S18][S19] A canonical core replaces bilateral adapters with a hub-and-spoke contract grounded in IEC 62264-1 (ISA-95) enterprise-control models and GS1 EPCIS visibility event structures.[S05][S08][S10] Each capture source maps once into the canonical model; each downstream product reads from that core.[S05][S10]
The divergence between bilateral translation and a canonical core across those four illustrative stages is shown in Exhibit 21.[S05][S10][S39]
As Exhibit 21 demonstrates, the payoff of a canonical core is modest with one source and one report, but compounds rapidly once shop-floor records serve operations, quality, maintenance, sustainability and machine learning.[S05][S10][S39] However, Exhibit 21 cannot prove that an expansive enterprise model is desirable; designing a schema for hundreds of theoretical entities before proving shop-floor value stalls implementation.[S16][S35][S40]
To avoid premature complexity, a small team must restrict the canonical core to cross-domain records carrying repeated decision value.[S05][S08][S38] Rather than mapping every internal controller register into the central model, the spine standardises only records crossing domain boundaries: site and line hierarchy, order and lot identity, shift calendar, output event, state transition (downtime) event, quality inspection and defect event, maintenance event, and utility consumption interval.[S05][S08][S38] High-frequency internal diagnostic signals that matter only to one machine's local maintenance remain at the edge or in the raw log until a validated use case requires promotion.[S06][S16][S33]
Interlude · the cloth half-woven
Woven: sections 5.0 to 5.7Still to weave: 5.10 to 5.19
Read-out
Inherited GFT programme evidence base prepared 31 January 2024. R06 and R07 were not refreshed in 2026 and their original publication dates are unrecorded. Figures describe sector-level baseline and policy ambition, not a current GFT factory measurement or causal productivity claim.
Unit: percent of cotton fibre. Status: sourced. Sources: R06, R07.
Read this exhibit in its chapter ↗5.10
Exhibit 227 paragraphs step it
Read-out
Derived store-and-forward governance flow combining local edge buffering during disconnection from S33 with transport delivery from S07, event identity and dual timestamps from S08, and provenance state tracking from S09.
Unit: . Status: derived. Sources: S07, S08, S09, S33.
Moving records from capture nodes to the raw store and canonical core requires transport suited to factory networks, but transport selection must never be confused with governance.[S06][S07] Teams often equate installing a message broker with building a data architecture, assuming that once payloads flow across a publish-subscribe bus, the factory has a usable spine.[S07][S10] That confusion leaves consumers parsing undocumented strings, guessing units, and misinterpreting retried messages as duplicate output.[S04][S07][S08]
Where multiple services need timely access to shop-floor events without polling controllers repeatedly, a publish-subscribe transport protocol such as MQTT Version 5.0 is an effective choice.[S07] A line-side gateway or vision unit publishes an event once to a broker topic, and the raw writer, operational monitor and alerting service subscribe without placing extra connection load on the shop-floor controller.[S07][S22] Where richer machine-to-machine contextual interoperability and browseable object hierarchies are needed at the device boundary, OPC Unified Architecture (OPC UA) provides an information-modelling and secure communication framework.[S06] For low-frequency shift summaries or batch test results, transactional hypertext transfer or scheduled file ingestion with checksums is often simpler for a small team to maintain than a streaming broker.[S16][S20]
When network links between a production hall and central storage drop, the transport contract must coordinate with local edge buffering, as traced in Exhibit 22.[S07][S08][S09][S33]
As Exhibit 22 shows, surviving a network outage requires six steps across capture, transport and ingestion.[S07][S08][S09][S33] First, the capture node assigns a globally unique event identifier and records observation time before transmission.[S08] Second, the event is written to a persistent local queue on the edge parent device so a link failure does not drop the record.[S33] Third, when connectivity returns, the transport client replays buffered messages in order.[S07][S33] Fourth, because at-least-once delivery can send a message more than once during reconnection, the receiver checks the event identifier for idempotency and stamps an ingestion timestamp before appending the unique payload to the raw store.[S07][S08] Fifth, the schema and quality gate validates the payload and routes failures to quarantine.[S04] Sixth, validated records advance to the canonical core with provenance showing delayed replay rather than live streaming.[S09] Note that Exhibit 22 covers data continuity during communication loss; it does not provide backup power or machine safety interlocking.[S22][S33]
OASIS defines MQTT Version 5.0 strictly as a client-server publish-subscribe messaging transport protocol.[S07] The protocol transports bytes between publishers and subscribers according to configured delivery levels, session expiry intervals and topic filters, without knowing whether a payload labelled line1/output represents finished first-quality garments, total pieces including rework, or cumulative counter ticks since a controller reboot.[S07][S38]
Treating broker throughput or topic hierarchy as proof of semantic governance is a recurring error.[S06][S07][S19] Every transport topic or ingestion endpoint must be bound to a versioned data contract specifying the schema, unit, time grain, event semantics, quality state rules and accountable owner.[S02][S04][S08] If a publisher alters its payload structure or unit without an approved contract version update, the validator quarantines the message even though the broker delivered the packet without error.[S04][S07]
5.13
The third separation stands between the canonical core and the people or systems that consume data.[S05][S0502] Granting analysts, dashboard authors, external auditors or scripts direct credentials to query raw ingestion tables or mutate canonical tables causes two failures: ad hoc queries diverge from the governed KPI dictionary, resurrecting conflicting figures for output, downtime and defects; and unthrottled queries or accidental writes contend with live ingestion and violate least privilege.[S18][S22][S25][S38]
In GFT's minimum spine, consumers never receive write access to raw evidence or canonical operational tables, and ordinary consumers do not query raw payloads directly.[S09][S25] Instead, the spine publishes controlled, read-only data products tailored to distinct consumption patterns, following W3C Data on the Web Best Practices for consistent dataset publication and the Data Quality Vocabulary principle of exposing fitness-for-purpose metadata alongside published datasets.[S0502][S0503] Access to each published product is governed by least privilege under NIST SP 800-53 Revision 5, granting each human role or automated service account only the permissions required to accomplish its assigned task.[S25]
Read-out
| 1. Shift operational view | Line supervisors and plant manager | Hourly and shift aggregates | Read-only by operational role | Uses active ISO 22400-2 KPI version; flags missing or buffered intervals |
|---|---|---|---|---|
| 2. Quality and steward queue | Domain stewards and quality leads | Event-level exceptions and quarantine | Read queue; append correction event | Links every superseding correction to raw event ID and steward role |
| 3. ML training snapshot | Model builder and validation pipeline | Frozen point-in-time dataset extract | Immutable read by pipeline identity | Records manifest hash, schema version, quality states and split window |
| 4. Traceability and carbon view | Sustainability and customer compliance | Lot, order and resource balances | Read-only scoped to external reporting | Joins EPCIS lot events with utility intervals; excludes worker data |
| 5. Raw diagnostic archive | Data custodian and incident responder | Unmodified ingestion payloads | Restricted privileged audit access | Append-only storage with cryptographic hashes and offline backup copy |
Derived product contract table mapping five factory consumption classes to their canonical inputs, refresh grain, access permissions and lineage guarantees, synthesised from S05, S08, S15, S25, S38 and S0502.
Unit: . Status: derived. Sources: S05, S08, S15, S25, S38, S0502.
As Exhibit 23 specifies, each consumer interacts with a bounded contract rather than an undifferentiated data store.[S15][S25][S0502] Line supervisors reviewing shift performance consume pre-aggregated views that implement the exact formula, time elements and units registered in the factory's ISO 22400-2 KPI dictionary, with visual badges whenever an hourly bucket contains buffered, missing or estimated records.[S04][S38][S0503] Machine-learning pipelines never train against live mutable tables; they consume frozen, content-hashed snapshots whose manifest records the query, canonical schema version and quality-rule version used to assemble the extract.[S04][S09][S15]
Publishing separate data products for operations, quality, sustainability and machine learning does not duplicate data pipelines.[S10][S39] Because every published product in Exhibit 23 is derived from the same canonical core and uses the same governed identifiers for site, line, asset, production order, material lot and shift, figures reconcile across departments by construction.[S05][S08][S38]
When the plant manager examines first-pass yield in the daily cadence (WP02), the quality engineer evaluates an optical model's false-rejection rate (WP03), the planner prepares an order forecast (WP04), the compliance lead traces a cotton bale to a garment lot (WP05), and the sustainability lead allocates kilowatt-hours and water per kilogram of fabric (WP16), all five decisions reference identical lot identifiers, order identifiers and validated output quantities.[S05][S08][S38] Fit-for-purpose transformations happen in the published product layer, such as converting piece counts into fabric mass for carbon accounting or lagging order histories into features for forecasting, while the underlying canonical facts remain single-sourced and versioned.[S09][S15][S38]
5.16
A disciplined library separates information governance from physical network and hardware engineering so neither concern is diluted. Within GFT's research architecture, WP14 owns the governance of the data spine: decision rights, operational vocabulary, data contracts, quality gates, provenance lineage, capture-context-consumption separation, recovery objectives and staged MLOps controls.[S01][S02][S09][S15] Companion paper WP18 owns the physical and logical reference topology: network segmentation, switch and firewall placement, edge compute sizing, container orchestration, storage engine selection and link engineering.[S06][S22][S28]
Separating WP14 from WP18 does not make them independent; governance sets the non-negotiable invariants that every topology choice in WP18 must satisfy.[S05][S10][S22] In small and medium-sized manufacturers, where the 2026 OECD Western Balkans survey illustrates that integrated tools such as enterprise resource planning (ERP), cloud computing and customer relationship management (CRM) are used by only 10 to 20 percent of surveyed firms (an external SME benchmark rather than a Cameroon prevalence measure), infrastructure teams face pressure to improvise quick hardware or cloud fixes.[S34] Without governance constraints from WP14, an infrastructure deployment can break record integrity, for example by configuring an edge gateway to overwrite local ring buffers silently without logging dropped intervals, or by deploying a database that lacks role-based separation between raw and curated tables.[S09][S25][S33]
Every topology and component design under WP18 must therefore pass six record-governance obligations defined here: 1. Boundary safety and non-interference: Network taps, polling gateways and edge inference nodes must respect NIST SP 800-82 Revision 3 operational technology performance, reliability and safety constraints, ensuring a data-spine fault cannot stall or alter machine control.[S22] 2. Disconnection durability: Any physical link between a capture cell and central storage must include durable local store-and-forward buffering sized for expected outages, preserving event identifiers and observation timestamps across link drops.[S08][S33] 3. Immutable raw separation: The storage engine and access-control configuration must enforce append-only retention for raw payloads, preventing in-place updates or unlogged deletions by normal users or transformation jobs.[S09][S25] 4. End-to-end provenance capture: Pipeline runners, brokers and schemas must record source agent, capture method, dual timestamps, schema version and transformation metadata from ingestion through serving.[S08][S09][S15] 5. Least-privilege product isolation: Authentication and authorisation infrastructure must support separate identities for operators, stewards, custodians, analysts, ingestion services and pipelines, restricting consumers to read-only published products.[S25] 6. Tested offline recovery and portable export: Backup and storage components must maintain offline, encrypted backup copies and support complete export of raw records, curated tables, schemas, KPI definitions and model artefacts in open formats that can be restored and verified independently.[S10][S26][S32]
To keep the boundary between WP14 and WP18 operational, the custodian and domain stewards verify infrastructure changes against record invariants before any new network segment, broker or storage node enters production.[S02][S27] When WP18 upgrades an edge gateway or migrates storage from a local server to a hybrid cloud instance, the acceptance test is not merely that packets cross the network or that a server reports healthy processor utilisation.[S17][S27] The acceptance test is whether a test batch of events injected at capture arrives in the raw store with intact hashes, passes or quarantines deterministically under the active contract, resolves against the canonical master-data version, reproduces historical KPI values in the operational view, and restores cleanly from an offline backup into an isolated test environment.[S04][S09][S26][S38]
5.19
Identifiers and KPI formulas
Identifiers and KPI formulas: Commodity tool availability 2, Need for internal semantic control 9.5. Quadrant: Own and author internally · illustrative
A small factory team cannot build and maintain custom database engines, message brokers or experiment-tracking platforms, nor should it surrender the meaning of its operations to an external vendor’s black-box schema .
Canonical event contracts
Canonical event contracts: Commodity tool availability 3.5, Need for internal semantic control 9. Quadrant: Own and author internally · illustrative
Quality rules and lineage graph
Quality rules and lineage graph: Commodity tool availability 5.5, Need for internal semantic control 8.2. Quadrant: Configure on open standards · illustrative
Model and experiment registry
Model and experiment registry: Commodity tool availability 6.8, Need for internal semantic control 6.5. Quadrant: Configure on open standards · illustrative
Dashboard and BI presentation
Dashboard and BI presentation: Commodity tool availability 7.5, Need for internal semantic control 5.5. Quadrant: Configure on open standards · illustrative
Message broker and transport
Message broker and transport: Commodity tool availability 8.8, Need for internal semantic control 2.5. Quadrant: Buy commodity utility · illustrative
Storage and encrypted backup
Storage and encrypted backup: Commodity tool availability 9.2, Need for internal semantic control 2. Quadrant: Buy commodity utility · illustrative
Illustrative architectural positioning of seven spine capabilities across commodity availability (x-axis, 0 = factory-specific to 10 = standardised commodity) and need for internal semantic control (y-axis, 0 = generic utility to 10 = core business meaning). Rationale derived from MLOps architecture in S20, connected-product data access considerations in S32 and internal champion ownership in S37.
Unit: illustrative score (0 to 10 scale). Status: illustrative. Sources: S20, S32, S37. Illustrative figures are teaching examples, not measured factory results.
5.19 · the bodyExhibit 24 above; its figures are illustrative where marked.
A small factory team cannot build and maintain custom database engines, message brokers or experiment-tracking platforms, nor should it surrender the meaning of its operations to an external vendor's black-box schema.[S16][S18][S32] Resolving the build-versus-buy decision requires disaggregating the spine by architectural layer rather than treating build and buy as an ideological choice across the factory.[S16][S20]
Transport brokers, relational or analytical storage engines, object storage, version control, backup encryption utilities and basic dashboard rendering are mature commodity capabilities.[S07][S16][S20] Writing custom software for message queuing, time-series compression or role-based authentication diverts scarce factory engineering hours into low-differentiation maintenance, compounding the hidden technical debt described in machine-learning systems literature.[S18][S20] In line with the Azure MLOps Maturity Model's guidance to develop capability gradually rather than facing full implementation complexity up front, small teams should buy or adopt proven, standard-compliant commodity software for transport, storage, backup and basic registry mechanics.[S16][S42] Equally, World Bank case study evidence on SME digital adoption in Viet Nam underscores that successful platform uptake depends on assigning a named internal point person who champions adoption within the firm; buying commodity software never removes the need for internal operational leadership.[S37]
Conversely, a factory must never outsource its operational semantics.[S05][S38] The identifier taxonomy (how sites, lines, machines, cotton lots, garment styles and shifts are coded), canonical event schemas, ISO 22400-2 KPI formulas, quality-gate thresholds and provenance lineage graph are the core of the factory's management system.[S04][S08][S09][S38] If a vendor's turnkey dashboard or vision appliance computes OEE or defect rates using proprietary internal definitions and refuses to expose underlying event records, the factory cannot reconcile those figures with its order ledger, cannot audit changes after firmware updates, and loses its historical baseline if it switches suppliers.[S10][S32][S38]
This layer-by-layer allocation of build, configure and buy choices is mapped in Exhibit 24.[S20][S32][S37]
As Exhibit 24 shows, capabilities in the lower-right quadrant (message brokers, storage, and encrypted backup utilities) combine high commodity availability with low need for bespoke business semantics, making them prime candidates to buy or adopt off the shelf.[S07][S20][S26] Capabilities in the upper-left quadrant (identifier schemes, canonical event contracts and KPI formulas) carry the factory's core operational meaning and must be authored and owned internally by the accountable domain owners and stewards.[S02][S08][S38] Capabilities in the upper-right quadrant (quality rule execution, lineage graphs, model registries and BI presentation) sit in the middle: the factory should adopt lightweight tools that support open standards, then configure them with its own governed definitions and metadata.[S09][S15][S20] Because Exhibit 24 is an illustrative structural framework rather than a cost survey, each site must evaluate vendor pricing, local support and staff skills through GFT's build-or-buy tool before signing a contract.[S37][S44]
Retaining semantic control when purchasing machinery, edge sensors or software platforms requires explicit procurement clauses before commissioning.[S10][S32] Where the European Union Data Act (Regulation (EU) 2023/2854) applies, for instance when a connected product or service falls within the territorial and material scope of Union rules, Recital 5 states that users in the Union can access, in a timely manner, the data generated by the use of that product or service and share those data with third parties of their choice.[S32] For a factory in Cameroon, the EU Data Act is not domestic law and must not be cited as an automatic statutory right; instead, it serves as a practical procurement benchmark for the access clauses the factory should write into its commercial equipment and software contracts.[S32]
Specifically, every procurement specification for machinery, optical inspection units, meters or cloud analytics in the spine should require four contractual deliverables:
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.
39 sources · 334 citation marks on this pageMost cited: S09, 27 times
Grade AGrade BGrade CKnot size: citations on this page
All 39 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.
R12 · Grade C · inherited record
Inherited evidence R12
CICAM capacity utilisation, July 2023 report: Garoua 33.7 percent, CICAM 1 48.9 percent, CICAM 2 10.83 percent; underutilisation cost about CFAF 2 billion in 2020.
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.
c05-S01 · Grade A
ISO 8000-1:2022: Data quality, Part 1: Overview
The series defines which characteristics of data are relevant to data quality, specifies requirements applicable to those characteristics, and provides guidelines for improving data quality.
2022. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S02 · Grade A
ISO 8000-150:2022: Data quality, Part 150: Data quality management: Roles and responsibilities
This document specifies the key considerations for organizations that are establishing appropriate roles and responsibilities for data quality management.
2022. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S03 · Grade A
ISO/IEC 25012:2008: Software engineering, Software product Quality Requirements and Evaluation, Data quality model
Data quality requirements and data quality measures can be categorized with the data quality characteristics in 5.2 and used by an evaluation process to analyse data independently from other computer system's components.
2008. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S04 · Grade A
ISO/IEC 5259-3:2024: Artificial intelligence, Data quality for analytics and machine learning, Part 3: Data quality management requirements and guidelines
ISO/IEC 5259-3 defines the requirements and guidance for managing the quality of data used in analytics and machine learning (ML).
2024. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S05 · Grade A
IEC 62264-1:2013, Edition 2.0: Enterprise-control system integration, Part 1: Models and terminology
This part of IEC 62264 provides standard models and terminology for describing the interfaces between the business systems of an enterprise and its manufacturing operations and control systems.
2013. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S06 · Grade A
OPC Unified Architecture, Part 1: Overview and Concepts, OPC 10000-1, version 1.05.06
It provides a comprehensive framework for modelling information, secure communication, and ensuring interoperability between diverse systems and devices.
2025-10-22. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S07 · Grade A
MQTT Version 5.0
MQTT is a Client Server publish/subscribe messaging transport protocol.
2019-03-07. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S08 · Grade A
EPCIS Standard, Release 2.0
The goal of EPCIS is to enable disparate applications to create and share visibility event data, both within and across enterprises.
2022. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S09 · Grade A
PROV-O: The PROV Ontology
It provides a set of classes, properties, and restrictions that can be used to represent and interchange provenance information generated in different systems and under different contexts.
2013-04-30. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S10 · Grade A
Current Standards Landscape for Smart Manufacturing Systems, NISTIR 8107
Standards are fundamental 1) to facilitate the delivery of the right information at the right time, 2) to enable actions based on that information and 3) to reduce risk of technology adoption and development.
2016-02-23. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S15 · Grade B
MLOps: Continuous delivery and automation pipelines in machine learning
Information about each execution of the ML pipeline is recorded in order to help with data and artifacts lineage, reproducibility, and comparisons.
2024-08-28. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S16 · Grade B
MLOps Maturity Model
Use this model to develop your MLOps capability gradually instead of facing the full complexity of mature implementation upfront.
2026-06-03. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S17 · Grade B
Model monitoring in production
You can set thresholds for these metrics to trigger alerts about model or data anomalies via Azure Machine Learning or Azure Event Grid.
2026-01-27. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S18 · Grade B
Hidden Technical Debt in Machine Learning Systems
Using the software engineering framework of technical debt, we find it is common to incur massive ongoing maintenance costs in real-world ML systems.
2015. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S19 · Grade B
Everyone wants to do the model work, not the data work: Data Cascades in High-Stakes AI
Data cascades are pervasive (92% prevalence), invisible, delayed, but often avoidable.
2021. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S20 · Grade B
Machine Learning Operations: Overview, Definition, and Architecture
MLOps includes several aspects, such as best practices, sets of concepts, and development culture.
2022-05-14. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S22 · Grade A
Guide to Operational Technology Security, NIST SP 800-82 Revision 3
This document provides guidance on how to secure operational technology (OT) while addressing their unique performance, reliability, and safety requirements.
2023. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S25 · Grade A
Security and Privacy Controls for Information Systems and Organizations, NIST SP 800-53 Revision 5
Employ the principle of least privilege, allowing only authorized accesses for users (or processes acting on behalf of users) that are necessary to accomplish assigned organizational tasks.
2020-12-10. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S26 · Grade A
#StopRansomware Guide
Maintain offline, encrypted backups of critical data, and regularly test the availability and integrity of backups in a disaster recovery scenario [CPG 2.R].
2023. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S27 · Grade A
Incident Response Recommendations and Considerations for Cybersecurity Risk Management: A CSF 2.0 Community Profile, NIST SP 800-61 Revision 3
Procedures can be tested or exercised periodically to verify their accuracy and can be used to help train new personnel.
2025. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S28 · Grade A
ISA/IEC 62443 Series of Standards
The ISA/IEC 62443 series addresses the security of industrial automation and control systems (IACS) throughout their lifecycle (which applies to all automation and control systems, not only industrial).
2026. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S29 · Grade A
Regulation (EU) 2016/679, General Data Protection Regulation
The personal data should be adequate, relevant and limited to what is necessary for the purposes for which they are processed.
2016-04-27. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S32 · Grade A
Regulation (EU) 2023/2854 on harmonised rules on fair access to and use of data, Data Act
This Regulation ensures that users of a connected product or related service in the Union can access, in a timely manner, the data generated by the use of that connected product or related service and that those users can use the data, including by sharing them with third parties of their choice.
2023-12-13. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S33 · Grade B
Understand extended offline capabilities for IoT Edge devices, modules, and child devices
If your devices go offline, the IoT Edge parent device stores all device-to-cloud messages until the connection is reestablished.
2024-06-10. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S34 · Grade B
Uptake of digital tools among SMEs: Western Balkans Enterprise Survey
Tools that support more integrated business operations, such as cloud computing, enterprise resource planning (ERP) and customer relationship management (CRM) systems, are used by between 10-20% of surveyed firms, while e-commerce and supplier-customer management systems are less common.
2026-06-26. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S35 · Grade B
Employee empowerment through digital tools: The case of a German bakery SME
However, following managing director Holste the change of traditional processes into digital ones requires sizeable efforts that do not directly pay off but involve a longer learning process.
2022-11-21. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S37 · Grade B
Challenges with digital technology adoption by SMEs: A case study from Viet Nam
Firms that successfully took up the platform assigned a point person who championed adoption within the firm.
2025-05-20. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S38 · Grade A
ISO 22400-2:2014: Automation systems and integration, Key performance indicators 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 ↗c05-S39 · Grade A
Artificial Intelligence with Open and Scaled Data Sharing in Semiconductor Manufacturing: Workshop Report
Also, manufacturers have reported that the cost of data processing for a single application can be as high as 70% of the total cost.
2025. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S40 · Grade B
The Digitization of TVET and Skills Systems
We suggest that digital transformation strategies have a better chance of success if they are developed to support tangible pilots that can demonstrate the practical use of emerging technologies with a short timeframe (two to three years maximum) and inbuilt mechanisms to deploy technology that provides ongoing data on what works and what does not to the sponsoring stakeholders.
2020. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S42 · Grade B
Is MLOps different in Industry 4.0? General and Specific Challenges
Also the starting-level of the technical infrastructure as well as the aimed at degree of automation influence very significantly the overall complexity of implementing MLOps in Industry 4.0.
2022. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S44 · Grade B
Supporting SME companies in mapping out AI potential: a Finnish AI development case
One must remember that in IT investments, the decision to postpone, e.g. an AI project, is also valid if resources, skills, costs, or expected benefits are not met.
2024-07-27. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S0501 · Grade A
PROV-DM: The PROV Data Model
Provenance is information about entities, activities, and people involved in producing a piece of data or thing, which can be used to form assessments about its quality, reliability or trustworthiness.
2013-04-30. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S0502 · Grade A
Data on the Web Best Practices
This document provides Best Practices related to the publication and usage of data on the Web designed to help support a self-sustaining ecosystem.
2017-01-31. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c05-S0503 · Grade A
Data Quality Vocabulary (DQV)
It does not provide a formal, complete definition of quality, rather, it sets out a consistent means by which information can be provided such that a potential user of a dataset can make his/her own judgment about its fitness for purpose.
2016-12-15. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗S05 International Electrotechnical Commission. 2013.
S10 National Institute of Standards and Technology. 2016-02-23.
R06 Cameroon Trade Hub. Publication date unrecorded; inherited source-text check.
R07 Ecofin Agency. Publication date unrecorded; inherited source-text check.
R12 Business in Cameroon. Publication date unrecorded; inherited source-text check.
S18 Advances in Neural Information Processing Systems. 2015.
S39 National Institute of Standards and Technology. 2025.
S08 GS1 AISBL. 2022.
S09 World Wide Web Consortium. 2013-04-30.
S15 Google Cloud. 2024-08-28.
S22 National Institute of Standards and Technology. 2023.
S33 Microsoft. 2024-06-10.
S19 Association for Computing Machinery, CHI 2021. 2021.
R29 MDPI; industry inspection literature. Publication date unrecorded; inherited source-text check.
S0501 World Wide Web Consortium. 2013-04-30.
S04 ISO/IEC. 2024.
S02 International Organization for Standardization. 2022.
S29 European Parliament and Council of the European Union. 2016-04-27.
S06 OPC Foundation. 2025-10-22.
S38 International Organization for Standardization. 2014.
S26 CISA, MS-ISAC, NSA and FBI. 2023.
S25 National Institute of Standards and Technology. 2020-12-10.
S01 International Organization for Standardization. 2022.
S0502 World Wide Web Consortium. 2017-01-31.
S03 ISO/IEC. 2008.
S0503 World Wide Web Consortium. 2016-12-15.
S16 Microsoft Learn, Azure Architecture Center. 2026-06-03.
S35 OECD. 2022-11-21.
S40 International Labour Organization and UNESCO. 2020.
S07 OASIS Open. 2019-03-07.
S20 arXiv, author version of an IEEE Access article. 2022-05-14.
S28 International Society of Automation. 2026.
S34 OECD. 2026-06-26.
S32 European Parliament and Council of the European Union. 2023-12-13.
S27 National Institute of Standards and Technology. 2025.
S17 Microsoft Learn, Azure Machine Learning. 2026-01-27.
S42 SCITEPRESS, IN4PL 2022. 2022.
S37 World Bank. 2025-05-20.
S44 The Journal of Technology Transfer, Springer Nature. 2024-07-27.
Keep the thread · chapter 05 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: Govern meaning before scale
The binding constraint on industrial AI is often the broken chain between work, record and decision, not the absence of an algorithm.
02Part I: Govern meaning before scale
A workable governance model fits into existing factory roles and makes every definition, exception and release somebody's explicit decision.
03Part I: Govern meaning before scale
Shared identifiers, event semantics and KPI formulas let unlike systems describe the same work without hiding their differences.
04Part I: Govern meaning before scale
A small set of enforced checks, visible exceptions and complete lineage is more useful than an unmeasured claim that the data are clean.
05Part II: Build the minimum durable spine
A durable architecture preserves source evidence, adds governed context once and lets each decision consume a controlled product.
06Part II: Build the minimum durable spine
A value becomes useful evidence only when the factory can establish what it describes, when it happened and which state or transformation it represents.
07Part II: Build the minimum durable spine
The data spine must protect production, people and legitimate access at the same time, with controls proportionate to a small team.
08Part II: Build the minimum durable spine
A factory proves resilience by operating through disconnection, restoring its records and leaving a vendor without losing its history.
09Part III: Control models as production change
The first model gate is operational and economic: nobody should train before the action, baseline, data rights and stop conditions are explicit.
10Part III: Control models as production change
A model becomes a release candidate only when another competent person can reconstruct what data, code, configuration and evaluation produced it.
11Part III: Control models as production change
Release authority follows demonstrated local fitness, safe fallback and accountable acceptance, not a strong aggregate validation score.
12Part III: Control models as production change
A useful monitor detects change, establishes whether it matters and routes a named person through a rehearsed response.
13Part IV: Prove value and scale deliberately
The spine earns further investment when people use it, decisions improve and net operational benefit survives a comparison with the baseline.
14Part IV: Prove value and scale deliberately
GFT should scale the spine only as ownership, meaning, recovery, reproducibility and net value are demonstrated in production.