Research libraryThe Industrial Data Spine
Chapter 06 of 14 · Part II · Draft web edition
A value becomes useful evidence only when the factory can establish what it describes, when it happened and which state or transformation it represents.
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
Six illustrative timeline entries for events A, B, C and D, one transport retry and one KPI recomputation. All times are invented to demonstrate the method. Dual-time semantics derive from S08 and S0601, and buffering from S33. The exhibit does not set a recommended lateness threshold.
Unit: minutes after 09:00 (illustrative). Status: illustrative. Sources: S08, S33, S0601. Illustrative figures are teaching examples, not measured factory results.
A sensor reading becomes industrial evidence only when it is bound to an event time, a stable operational identity and an explicit state or transformation. A counter value alone cannot establish whether output occurred, which asset or lot produced it, whether the line was running, or whether a later replay has already counted it. The chapter's answer is therefore a governed evidence chain: keep event time separate from ingestion time, resolve events against versioned master data, represent genealogy and state transitions explicitly, make replay idempotent, append corrections, and retain each record class for a stated purpose.[S05][S08][S09] The sensor value remains an observation. Evidence is the reconstructable account of what that observation meant when a person, KPI or model used it.
This distinction lets a factory migrate from manual hourly logs to machine ingestion without rewriting its operating history. In Cameroon, where GFT's programme evidence base 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 (with original source dates unrecorded and not refreshed in 2026), expanding domestic processing depends on operational records that hold up across shifts, power interruptions and buyer audits.[R06][R07] Preserving time, identity and state prevents a delayed connection from masquerading as a production stop, a renamed asset from losing its past, a split lot from acquiring a fictitious single batch identity, and a retried message from becoming additional output.[S08][S33][S38] It also gives later forecasting and inspection models the same temporal and operational definitions used on the floor, rather than allowing each training pipeline to invent them again.[S04][S15]
6.1
Related · chapter 049 paragraphs step it
Read-out
Derived operational control flow linking raw capture, entry syntax check, ingestion contract gate, quarantine isolation, steward remediation and provenance-logged reprocessing, based on ISO 8000-1 (S01), ISO/IEC 5259-3 (S04) and W3C PROV-O (S09).
Unit: . Status: derived. Sources: S01, S04, S09.
Read this exhibit in its chapter ↗Every governed event needs two timestamps. Event time records when the represented operational occurrence happened. Ingestion time records when the receiving system accepted its record. EPCIS makes the equivalent distinction explicit: its recordTime is a bookkeeping mechanism and does not describe the real-world event.[S0601] A factory should also distinguish processing time, when a particular transformation ran, from both. A stoppage that occurred during one shift but arrived during the next belongs to the first shift's operating account and to the second shift's data-latency account. Those are different questions, and neither timestamp can answer both.[S08][S33]
The raw record should preserve the source's event timestamp, its stated time-zone offset and the receiving system's ingestion timestamp without overwriting either. Normalising an additional representation to Coordinated Universal Time makes cross-system ordering consistent, but the original local representation remains useful evidence about what the source asserted.[S08] Ingestion time supports freshness monitoring, replay diagnosis and reconstruction of what a dashboard could have known at a given moment. Event time supports output windows, state intervals, genealogy and features that must not use future information.[S15][S38]
This separation changes operational behaviour. When a gateway reconnects, the custodian investigates the interval between event and ingestion timestamps as a continuity issue. The production steward does not automatically treat that interval as missing output or unplanned downtime. Conversely, a feed can arrive promptly with a wrong event timestamp because a device clock has drifted. Low ingestion latency therefore proves delivery speed, not temporal accuracy.[S0601][S33]
GFT's event contract adds clock source, observed clock offset or uncertainty, and a source sequence number where the device provides one. These fields are engineering metadata, not decorative compliance. A line-side terminal, programmable controller and manual back-entry form can assert different degrees of temporal confidence even when their displayed timestamps have the same precision. Recording the source and an estimated uncertainty lets a steward decide whether two events can be ordered safely or need review.[S08][S09]
The estimate needs a method. At commissioning and after a gateway restart, the custodian can compare the device with the site's chosen time reference, record the observed offset and retain the check date. Any uncertainty attached to a record should be labelled an estimate based on that check, not presented as a measured guarantee for every later event. Where the uncertainty crosses an hourly or shift boundary, the record should carry a boundary-review flag rather than a falsely precise allocation. A source sequence can help diagnose order within one device, but it cannot establish order across independent devices and cannot replace event time.[S05][S0601]
Manual entries require the same honesty. If a supervisor records a stoppage after the event, the form should ask when it happened and separately capture when the entry was submitted. It should not silently replace the first with the second. Back-entry status and the submitting role become provenance, while any unnecessary worker identity remains outside the production feed.[S09][S29]
An hourly bucket is a governed calculation, not a natural property of incoming messages. The KPI contract must state its event-time window, inclusion rule, provisional period, late-event treatment and formula version, consistent with ISO 22400's emphasis on formula, elements, time behaviour and unit.[S38] A live operational view may publish a provisional total so the shift can act, then issue a new calculation version when an accepted late event arrives. A monthly report may use a longer closure period. Neither waiting period is a universal constant; each is a site decision based on decision urgency and observed delay.
The same rule must reach model datasets. If a training feature describes an hour, its manifest should show whether late events were included up to a defined ingestion cut-off and which recalculation version produced the value. Otherwise a model can learn from a more complete history than will exist when it makes a live prediction.[S04][S15] Exhibit 25 isolates the temporal mechanism before the worked example adds quantities.
The important result in Exhibit 25 is not the arrival order. It is the preservation of two valid histories: what happened in production, and what the information system knew at each point. C does not move to D's hour merely because it arrived after reconnection, and the retry of A does not create a new production event. The exhibit cannot tell a factory how long to keep a bucket provisional, nor can it prove that the source clocks were correct. Those require site rules and clock evidence.[S0601]
6.5
An event identifier says which record is being described. A master identifier says which persistent object the record concerns. The second remains stable while names, locations, assignments and other attributes change. IEC 62264 supplies role-based equipment and physical-asset hierarchies, while OPC UA provides an information-modelling framework for objects and their relationships; neither removes the factory's duty to decide which record is authoritative.[S05][S06] GFT therefore separates the slow-changing master record from the high-frequency event stream and gives every master class an identified origin, steward and effective history.
A sewing asset can keep one immutable asset identity when its display name changes, when it moves to another line, or when its controller is replaced. Those are three different changes. Renaming alters an attribute. Moving alters a role-based assignment with a start and end time. Replacing the controller may alter a technical relationship without replacing the production asset, provided the asset owner confirms that operational continuity remains valid.[S05][S06] The event never carries a mutable display name as its only link to the object.
Each derivation then resolves the master version effective at event time, not whichever version happens to be current when a query runs. A historical output total remains assigned to the line on which the asset operated at that time. A current capacity view uses the current assignment. Both views can coexist because they state their temporal rule. The curated record should retain the master-data version or effective-dating evidence used in the join, so a later hierarchy edit cannot silently redistribute history.[S05][S09]
Master data also need a merge and retirement rule. If two registers contain the same physical asset, the steward does not delete one and leave dangling events. The registry records a mapping from the superseded identifier to the surviving identifier, the approval, the effective time and the reason. Existing raw events retain their original identifier; curated products apply the governed mapping and preserve that transformation in lineage.[S02][S09] An identifier that has been retired must not be issued to a different object, because reuse would make old events indistinguishable from new ones.
Exhibit 26 turns these rules into a compact registry design. Its entries are illustrative classes and controls, not a claim that a particular factory already operates this registry.
Read-out
| Asset | Engineering asset register | Immutable asset ID; dated line and controller links | Events, maintenance and resource allocation |
|---|---|---|---|
| Order | Order-management record | Order ID with versioned style, quantity and due attributes | Scheduling, output and delivery reconciliation |
| Lot | Material receipt or transformation event | Lot ID plus parent and child event links | Genealogy, quality and provenance |
| Style | Product-definition register | Style ID with dated operation and unit versions | Routing, costing and feature derivation |
| Operation | Versioned routing definition | Operation ID effective within a style version | Work-centre events and cycle-time analysis |
| Defect code | Quality taxonomy owner | Stable code with dated label and mapping history | Inspection, rework and model labels |
| Worker role | Authorised role register | Role ID only; no person ID in production events | Authority checks and shift-level aggregation |
Illustrative master-identity registry based on equipment and enterprise boundaries in S05, object and relationship modelling in S06, and personal-data minimisation in S29. Rows are recommended design examples, not observed implementations or GFT operating results.
Unit: illustrative registry design. Status: illustrative. Sources: S05, S06, S29. Illustrative figures are teaching examples, not measured factory results.
The registry in Exhibit 26 deliberately gives worker role a different treatment from machine and material identities. A production event may need to show which role was authorised to declare a transition, without attaching a persistent person identifier to every output record.[S29] The table also shows why master data cannot be reduced to a list of codes: each class has an origin, an effective history and a permitted use. What it does not supply is the site-specific code format, which should follow existing operating needs rather than an imposed naming convention.
6.8
Genealogy is a chain of event relationships, not a batch field copied from one database row to the next. EPCIS is designed to share visibility event data about objects in a business context, while PROV-O represents entities, the activities that transform them and the agents responsible.[S08][S09] Together they support a practical factory model: receipt creates an identified input lot, splitting relates children to their parent, joining records all inputs to a new output, and rework creates a traceable transformation rather than silently changing the original lot's status.
A single batch_id on a garment order fails as soon as production has more than one input or output. If a cotton lot is split, each child needs its own identity and a link to the parent. If yarn from two lots enters one fabric batch, the output needs links to both. The graph must be many-to-many, with the transformation event carrying the event time, location, operation and responsible role needed to interpret the relationship.[S08][S09] Copying only the most recent lot identifier would discard the earlier inputs and make provenance dependent on memory.
Rework needs equal care. The original inspection event and rejected state remain evidence. A rework event identifies the affected object, operation and outcome; a subsequent inspection creates a new disposition. If the physical item retains its identity, the events carry its changing state. If the process creates a new roll, lot or bundle, the new identifier receives a derivation link. The choice follows physical practice, not database convenience.[S08][S09]
Exhibit 27 shows the minimum relationship pattern. It is a derived event design, not a mass balance and not an observed production history.
Read-out
Derived genealogy pattern using visibility events from S08 and entity, activity and derivation relationships from S09. Identifiers are illustrative and links carry no quantities. The diagram does not demonstrate material conservation, legal chain of custody or carbon allocation.
Unit: event and entity relationships. Status: derived. Sources: S08, S09.
The flow in Exhibit 27 preserves why illustrative order O has a more complicated ancestry than one lot code can express. It also separates an inspection observation from the transformation that followed it. The links must carry quantities in the actual event records where allocation requires them, but the diagram deliberately makes no conservation claim. WP05 owns provenance and chain-of-custody method; WP16 owns carbon allocation. This chapter supplies the temporal and identity evidence those methods need, not their accounting conclusions.[S08][S09]
6.10
Time: running
Time: running. From Time: planned stop (Authorised restart), Time: unplanned fault (Repair verified), Time: blocked (Downstream released), Time: starved (Input available). Leads to Time: planned stop (Authorised stop), Time: unplanned fault, Time: blocked, Time: starved. · illustrative
A state is a governed interpretation of an entity during an interval or after an event, not a free-text label on a dashboard.
Time: planned stop
Time: planned stop. From Time: running (Authorised stop). Leads to Time: running (Authorised restart). · illustrative
Time: unplanned fault
Time: unplanned fault. From Time: running. Leads to Time: running (Repair verified). · illustrative
Time: starved
Time: starved. From Time: running. Leads to Time: running (Input available). · illustrative
Unit: first inspection
Unit: first inspection. Starting point. Leads to Unit: accepted or scrapped (Accept or scrap directly), Unit: rework (Repairable defect). · illustrative
Unit: rework
Unit: rework. From Unit: first inspection (Repairable defect). Leads to Unit: accepted or scrapped (Reinspection decides). · illustrative
Unit: accepted or scrapped
Unit: accepted or scrapped. From Unit: first inspection (Accept or scrap directly), Unit: rework (Reinspection decides). End point. · illustrative
Illustrative asset-time and unit-disposition state machines. Transition design and the requirement to state KPI time elements derive from S38. Reason codes, transition authority and acceptance thresholds remain site decisions. The two tracks must not be added as if they had a common denominator.
Unit: illustrative state transitions. Status: illustrative. Sources: S38. Illustrative figures are teaching examples, not measured factory results.
6.10 · the bodyExhibit 28 above; its figures are illustrative where marked.
A state is a governed interpretation of an entity during an interval or after an event, not a free-text label on a dashboard. Machine-time states and unit-disposition states answer different questions. A sewing line can be running while a garment is in rework, and a line can be stopped while an operator inspects a unit. Treating rework as a machine state or blocked as a defect disposition mixes those dimensions and makes downtime and defect totals irreconcilable.[S38] GFT therefore keeps separate state machines for asset time and material disposition, linked by events but never added together as if they shared a denominator.
Each state definition needs an entity class, meaning, entry condition, exit condition, permitted reason codes, overlaps and authoritative role. The event that changes state records the previous and next values, event time, ingestion time, source and transitioning role. A supervisor may authorise a planned-stop classification, a maintenance role may verify restart after a fault, and a quality role may authorise rework or scrap. The small-team role assignments remain explicit rather than being buried in software permissions.[S02][S09]
Allowed overlaps must be equally precise. Under an exclusive asset-time model, one asset cannot be running and in an unplanned fault during the same interval. It can, however, be in an unplanned fault while an order is awaiting material and a unit is held for inspection, because those states belong to different entities and dimensions. Where management wants simultaneous technical and business states, the contract should publish separate fields rather than force both into one status column.[S38]
Late and back-entered transitions follow the same temporal rule. If a fault began before it was entered, its interval begins at the evidenced event time and its ingestion record shows the delay. A missing end transition leaves an open interval flagged for steward action, not an automatic zero-duration stop. If a later correction changes the start or end, the original transition remains available through the append-only correction mechanism described below.[S0601][S0602]
For each asset-time model, the base states should form a complete and non-overlapping partition of the time being measured. Reason codes add diagnosis without creating a second duration. A fault interval carries one fault reason; it does not also create a generic downtime interval that is added to the fault duration. Planned stop, blocked and starved intervals enter availability or other KPI calculations only through the exclusions and time elements in the active KPI version.[S38] The mapping is part of the definition, not an analyst's later preference.
Unit disposition follows a separate event path. First-pass acceptance, rework, reinspection and scrap describe the fate of a unit or lot. Defect counts need a stated unit and inspection boundary, just as downtime needs a stated asset and time boundary. A second inspection after rework must not silently become another first-pass unit. These controls make the KPI numerator and denominator reconstructable rather than merely reproducible as code.[S04][S38]
Exhibit 28 places the two state machines side by side. It illustrates transition logic and does not prescribe one universal KPI formula.
The upper logic in Exhibit 28 allocates asset time once. The lower logic tracks material disposition without creating another downtime category. A blocked interval and a starved interval can carry different operational causes while receiving the same treatment in a particular availability formula, but that formula must be cited by version.[S38] The exhibit cannot determine whether a planned stop should be excluded at a given factory, nor can it establish that a source system's transitions are complete. Acceptance requires interval reconciliation against a known shift calendar and sampled physical observations.
Interlude · the cloth half-woven
Woven: sections 6.0 to 6.10Still to weave: 6.13 to 6.19
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 ↗6.13
Naive message sum at 10:00
Naive message sum at 10:00: total 60. Running total 60 illustrative pieces in 09:00 bucket · illustrative
Buffered delivery and transport retries protect availability, but they also create a governance risk: the same application event can arrive more than once or after later events.
Remove duplicate retry A
Remove duplicate retry A: removes 20. Running total 40 illustrative pieces in 09:00 bucket · illustrative
Provisional 09:00 bucket
Provisional 09:00 bucket: total 40. Running total 40 illustrative pieces in 09:00 bucket · illustrative
Allocate late buffered C
Allocate late buffered C: adds 20. Running total 60 illustrative pieces in 09:00 bucket · illustrative
Recalculated 09:00 bucket
Recalculated 09:00 bucket: total 60. Running total 60 illustrative pieces in 09:00 bucket · illustrative
The method behind it: Method.
Illustrative reconciliation from the worked example: 3 arrived messages x 20 pieces = 60 naive pieces at the 10:00 cut, less 20 pieces for the duplicate retry of A (giving 40 provisional pieces), plus 20 pieces when buffered event C arrives at 10:07 and is allocated to its 09:42 event time (giving 60 recalculated pieces). Values are invented for method demonstration, not GFT operating results.
Unit: illustrative pieces in 09:00 bucket. Status: illustrative. Sources: S07, S33, S38, S0601, S0603. Illustrative figures are teaching examples, not measured factory results.
6.13 · the bodyExhibit 29 above; its figures are illustrative where marked.
Buffered delivery and transport retries protect availability, but they also create a governance risk: the same application event can arrive more than once or after later events. MQTT supplies publish-subscribe transport, not a factory-specific rule for recognising production identity.[S07] The capture boundary must therefore issue a stable event identifier that survives every retry and buffer cycle, such as a time-ordered RFC 9562 UUIDv7 value generated when the occurrence is recorded.[S0603] A repeated transmission keeps that identifier; a genuinely new occurrence receives a new one even if its value, asset and timestamp happen to match another record.[S08][S33]
such as a time-ordered RFC 9562 UUIDv7 value generated when the occurrence is recorded
S0603
Idempotency means that processing the same event again produces no additional unintended effect. The raw store accepts one identity once, and every derived consumer either tracks the event identifiers already applied or recomputes from a defined input set rather than blindly incrementing a total. A dataset manifest records the included identifiers and rule versions so a training snapshot can be rebuilt without double counting.[S04][S15] Deduplication is therefore a chain-wide obligation, not only a broker setting.
A conflict needs a different response from a duplicate. If two messages carry the same event identifier and the same canonical payload hash, the consumer can recognise a retry. If they carry the same identifier but different hashes, it must quarantine an identity conflict and preserve both payloads for investigation. Choosing the latest would allow a transmission fault or unauthorised change to rewrite evidence without explanation.[S04][S09]
A controlled replay reads the preserved raw set, applies a named schema, master-data, state and KPI version, then publishes a new derived version with lineage to the execution. It does not mutate an old product in place and pretend that consumers saw the same result all along.[S09][S15] Where the same inputs and rules produce a different output, the team has found nondeterminism or an unrecorded dependency. That defect should block acceptance until explained.
The worked example shows why replay acceptance must reconcile identities, event-time buckets and calculation versions together. Deduplication alone would still misallocate C. Event-time allocation alone would still count the retry. A versionless dashboard could calculate the correct final total while hiding the fact that its earlier figure had changed. Exhibit 29 summarises that reconciliation for the 09:00 hourly bucket across both adjustments. Operational monitoring should therefore retain provisional revision history, duplicate-conflict counts and lateness distributions alongside the published KPI.[S04][S38]
As Exhibit 29 makes visible, the naive message sum at 10:00 and the recalculated event-time bucket after 10:07 share the same illustrative value of 60 pieces, yet the first figure is wrong twice: it includes a duplicate retry of A and omits buffered event C.[S07][S33][S0601] Without identifier deduplication and dual-time accounting, those two errors mask each other in the 09:00 bucket while inflating the two-hour total to 100 illustrative pieces.[S38][S0603] What Exhibit 29 cannot establish is the site-specific outage distribution or the cut-off window for closing provisional shift reports, which must be set from measured plant connectivity.[S33]
6.16
A correction changes the best supported account of an event; it must not destroy the earlier assertion or the fact that people acted on it. EPCIS error declarations follow this principle directly: the prior event is not modified, and subsequent queries return both the prior event and the declaration.[S0602] In W3C PROV-O, a revision is modelled explicitly as a derivation in which the resulting entity is a revised version of an original entity.[S0604] GFT applies the same pattern across manual and machine records. A superseding event identifies what it corrects, states the reason and changed fields, and records the submitting and approving roles with their declaration times. The original raw payload remains untouched.[S02][S09]
A supervisor can discover today that yesterday's lot code was wrong. The corrected lot relationship takes effect for the current operational view, but the factory still needs to reconstruct yesterday's view when the original code was believed correct. Correction time records when the new assertion entered the system. Event time records the production occurrence being corrected. Keeping both prevents a present-day repair from pretending that past users possessed information they did not have.[S0601][S0602]
The curated model therefore supports two questions without confusing them. “What is our best current account?” follows the supersession chain to the effective record. “What did this report or model know at its execution?” uses the records and correction cut-off captured in its lineage. PROV-O derivation and revision relationships provide a standards-based way to represent those links, while the factory's own schema supplies reason, authority and approval detail.[S09][S15]
A correction can itself be wrong. The chain must allow a later superseding assertion with the same evidence requirements rather than editing the first correction. Current-state queries need a deterministic rule for selecting the effective tip of that chain, and any branching corrections require steward resolution. This avoids both destructive editing and an ambiguous collection of competing truths.
Every material correction triggers an impact assessment. The steward and custodian identify affected state intervals, genealogy paths, KPI windows, published products and dataset snapshots before releasing the new effective state. The pipeline then produces new calculation versions with links to the correction event and the exact rule versions used. A previously issued management report retains its original value and receives an addendum where the operating owner decides that restatement is required.[S02][S09][S38]
Frozen training data follow the same rule. The pipeline does not mutate an old snapshot and retain its identifier. It creates a new manifest that records corrected event identifiers, supersession links and the ingestion cut-off, then evaluates whether the change affects a candidate or deployed model.[S04][S15] This deliberate path distinguishes legitimate historical revision from silent drift and makes the operational consequence of a correction measurable.
6.19
Retention is a decision about each record class, not a promise to keep everything forever or delete everything on one schedule. Raw evidence may be needed to reconstruct a customer inquiry or diagnose a recurring fault. Curated events need their schemas and master versions. Labels, snapshots and model artefacts need the lineage required to explain historical outputs. Access logs serve security investigation and have a different exposure and purpose.[S09][S15][S29] Each class therefore receives a purpose, owner, review trigger, deletion method and, where applicable, contractual or legal basis.
The retention register should address the classes separately:
These are decision fields, not proposed retention durations. A small factory should set actual periods only after reviewing operational reuse, customer terms, litigation or incident holds, and applicable law. Indefinite retention is not a conservative default when it increases privacy exposure and storage burden without a stated decision need.[S29]
An executable retention procedure covers primary stores, replicas, local buffers, analytical exports and backup cycles. It records a minimal disposal receipt containing record class, scope, authority, execution time and verification result, not the deleted payload. Where immediate deletion from every backup is not feasible, the procedure isolates expired backups from ordinary use and ensures that a restoration cannot silently return deleted personal records to production. A legal or incident hold is scoped, approved and reviewed rather than becoming an invisible permanent exception.[S29][S31]
Aggregates require the same discipline. Removing direct identifiers does not automatically make a small-group or event-level dataset non-personal. The owner must obtain an appropriate assessment before treating an aggregate as exempt from the personal-data schedule. Where retention ends for raw operational evidence, derived records should not claim a lineage that can no longer be inspected; their metadata should state the disposal boundary honestly.[S09][S29]
Time, identity and state turn readings into evidence because they make every later use answerable to its original context. Replay, correction and retention then preserve that answerability as systems change. The factory can reconstruct not only a value, but also the object, occurrence, transformation, rule and knowledge state behind it.
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.
18 sources · 130 citation marks on this pageMost cited: S09, 19 times
Grade AGrade BGrade CKnot size: citations on this page
All 18 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.
c06-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 ↗c06-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 ↗c06-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 ↗c06-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 ↗c06-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 ↗c06-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 ↗c06-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 ↗c06-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 ↗c06-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. This requires, in particular, ensuring that the period for which the personal data are stored is limited to a strict minimum. Personal data should be processed only if the purpose of the processing could not reasonably be fulfilled by other means. In order to ensure that the personal data are not kept longer than necessary, time limits should be established by the controller for erasure or for a periodic review.
2016-04-27. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c06-S31 · Grade A
Law No. 2024/017 of 23 December 2024 relating to personal data protection in Cameroon
The processing of personal data shall be subject to prior authorization by the Personal Data Protection Authority.
2024-12-23. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c06-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 ↗c06-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 ↗c06-S0601 · Grade A
EPCIS Standard, Release 2.0, Section 7.4.1: EPCISEvent
The recordTime does not describe anything about the real-world event, but is rather a bookkeeping mechanism that plays a role in the interpretation of standing queries as specified in section 8.2.5.2.
2022. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c06-S0602 · Grade A
EPCIS Standard, Release 2.0, Error declarations
The prior event is not modified in any way, and subsequent queries will return both the prior event and the error declaration.
2022. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c06-S0603 · Grade A
RFC 9562: Universally Unique IDentifiers (UUIDs)
UUIDv7 features a time-ordered value field derived from the widely implemented and well-known Unix Epoch timestamp source, the number of milliseconds since midnight 1 Jan 1970 UTC, leap seconds excluded.
2024-05. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗c06-S0604 · Grade A
PROV-O: The PROV Ontology, Section 3.2: Expanded Terms
A revision is a derivation for which the resulting entity is a revised version of some original.
2013-04-30. Register access date: 2026-10-30. Source provenance is supplied by the paper evidence register.
Open the recorded source ↗S05 International Electrotechnical Commission. 2013.
S08 GS1 AISBL. 2022.
S09 World Wide Web Consortium. 2013-04-30.
R06 Cameroon Trade Hub. Publication date unrecorded; inherited source-text check.
R07 Ecofin Agency. Publication date unrecorded; inherited source-text check.
S33 Microsoft. 2024-06-10.
S38 International Organization for Standardization. 2014.
S04 ISO/IEC. 2024.
S15 Google Cloud. 2024-08-28.
S0601 GS1 AISBL. 2022.
S29 European Parliament and Council of the European Union. 2016-04-27.
S06 OPC Foundation. 2025-10-22.
S02 International Organization for Standardization. 2022.
S0602 GS1 AISBL. 2022.
S07 OASIS Open. 2019-03-07.
S0603 Internet Engineering Task Force. 2024-05.
S0604 World Wide Web Consortium. 2013-04-30.
S31 Presidency of the Republic of Cameroon. 2024-12-23.
Keep the thread · chapter 06 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.