Context card
Context cards
Short, self-contained explanations of the technical contexts behind the news: one question, a direct answer, the key points and the glossary terms it builds on. In US English, British English and German.
How to read a context cardRole card Functional role
Ontology steward: the functional role
Who keeps the governed semantic core of a DITA and iiRDS knowledge architecture stable while the model evolves?
The ontology steward keeps the governed ontologies stable and versioned and models approved changes; the function behind the semantic contract.
Ontology · Semantic model · Semantic versioning · Schema governance · RDF (Resource Description Framework)
Read the cardContext cardTransitional topic
What concept theory offers ontology work
What can ontology and knowledge graph work take from research on how concepts are represented?
Three lenses from concept theory for ontologies and knowledge graphs: definitional vs. typical, comparable values, and one occurrence on several levels.
Ontology · Knowledge graph · Information model · RDF (Resource Description Framework) · Semantic drift · Ontology alignment · Frame (concept representation) · Cascade (frame theory) · Conceptual space · Lexical decomposition · Conceptual atomism · Default inheritance (knowledge representation)
Read the cardContext cardTransitional topic
Decomposition or atomism: how much should a definition carry?
Does the meaning of a term have parts that a definition can list?
Decomposition vs. atomism: whether meanings have parts a definition can list — and what that means for glossaries, terminologies and ontologies.
Ontology · Semantic drift · Information model · Lexical decomposition · Conceptual atomism · Analytic–synthetic distinction
Read the cardContext card
Pause, abort and undo: three functions in agent execution
What is the difference between pausing, aborting and undoing an action an AI agent has started?
Pausing holds the next step, aborting ends the run, undoing needs the target system. Only one of the three takes a submitted form back.
Control (automation) · Transaction · Hypertext Transfer Protocol (HTTP) · Security (IT and OT)
Read the cardRole card Functional role
Content developer: the functional role
Who writes and maintains the content a DITA and iiRDS digital twin is built from?
The content developer writes the DITA topics and iiRDS metadata a knowledge graph is built from, and reports what readers ask that the content can’t answer.
DITA (Darwin Information Typing Architecture) · DITA topic · Metadata · iiRDS InformationType · User role
Read the cardRole card Functional role
Semantic governance approver: the functional role
Who decides which proposed changes to the ontology become part of the enterprise model?
The semantic governance approver decides which validated ontology changes reach the enterprise model and records why — never an agent, always people.
Schema governance · Data governance · Semantic model · Ontology
Read the cardContext card
Three processing purposes when AI supports case handling
Which processing purposes have to be kept apart when an AI system supports the handling of individual cases?
Model development, procedure analysis and support in a single case are three purposes with three data sets and three sets of duties.
Human oversight (AI Act) · Data quality · Validation · Technical documentation
Read the cardContext card
Agent fleet or agent swarm: what parallel activity shows
What is the difference between an agent fleet and an agent swarm, and which of the two can an outside observer establish?
A fleet is many agents in parallel; a swarm coordinates. Outside traces show requests and timing, not coordination.
Communication · Network · Security (IT and OT) · Technical documentation
Read the cardContext card
Why a copied interface is not a test boundary
What makes a test setup for an AI agent isolated — a copied interface or an enforced boundary?
A copied screen does not isolate an agent test. The boundary has to hold for tools, destinations, credentials and redirects.
Evaluation (evals) · Network · Software defect · Security (IT and OT)
Read the cardRole card Functional role
Information user: the functional role
Whose questions drive an adaptive semantic architecture — and what may the system do with them?
Information users ask the system for what they need to do a job; their recurring questions are the evidence the evolution cycle learns from.
AI agent · User role · Retrieval-augmented generation (RAG)
Read the cardRole card Functional role
Semantic platform operator: the functional role
Who runs the graphs, the inference, the semantic view and the vector index — and keeps them in step with the approved model?
The semantic platform operator runs the graphs, inference, the semantic view and the vector index, and keeps them in step with the approved model version.
Knowledge graph · Digital twin · Embedding · Vector database · Semantic versioning
Read the cardRole card Functional role
AI agent as proposer: the functional role
What is the role of the AI agent in an adaptive semantic architecture — tool, source of evidence or decision maker?
The AI agent consumes the semantic model and supplies evidence for its improvement; it may draft candidate changes, but never approves them.
AI agent · Knowledge graph · Retrieval-augmented generation (RAG) · Semantic model
Read the cardPathway map Map
From knowledge graphs to adaptive semantics: the pathway map
How does a DITA and iiRDS knowledge architecture get from governed semantics to a model that learns from AI agent use — and who does what on the way?
Michael Iantosca’s white paper on adaptive semantics as a pathway map: seven lanes, 21 stations with maturity, four role routes through governance.
DITA (Darwin Information Typing Architecture) · DITA topic · iiRDS InformationType · Ontology · Knowledge graph · Digital twin · Semantic model · Embedding · Vector database · AI agent · Schema governance · Semantic versioning
Read the cardContext card
ISO 20607 and IEC/IEEE 82079-1: two standards, one instruction handbook
Which standard should a machine builder follow for the instruction handbook — ISO 20607 or IEC/IEEE 82079-1?
ISO 20607 covers the safety-related content of a machine's handbook; IEC/IEEE 82079-1 the process and quality of information for use in general.
ISO 20607 (instruction handbook for machinery) · IEC/IEEE 82079-1 · ISO 12100 · Instructions for use (IFU) · Information for use · Operating manual · Warning message · Harmonized standard · Presumption of conformity · Target audience analysis · Machinery Regulation (EU) 2023/1230 · Digital instructions for use (Machinery Regulation)
Read the cardContext card
Digital instructions under the Machinery Regulation: what the law allows and what it still asks
May a machine builder supply the instructions for use only in digital form under the Machinery Regulation?
From 2027 the Machinery Regulation allows digital instructions — with access, download, ten-year availability and paper on request.
Digital instructions for use (Machinery Regulation) · Machinery Regulation (EU) 2023/1230 · Machinery Directive 2006/42/EC · Instructions for use (IFU) · Operating manual · ISO 20607 (instruction handbook for machinery) · IEC/IEEE 82079-1 · Information and instructions to the user (CRA) · VDI 2770
Read the cardContext card
Machine instructions and CRA user information: one information set, two sets of rules
What does the Cyber Resilience Act add to the instructions of a connected machine?
For a connected machine the CRA adds Annex II security information next to the machinery instructions — one information set can serve both sets of rules.
Information and instructions to the user (CRA) · Support period (CRA) · Cyber Resilience Act (CRA) · Machinery Regulation (EU) 2023/1230 · Digital instructions for use (Machinery Regulation) · ISO 20607 (instruction handbook for machinery) · IEC/IEEE 82079-1 · Coordinated vulnerability disclosure (CVD) · Software bill of materials (SBOM) · Update capability · Decommissioning · Instructions for use (IFU)
Read the cardPathway card Pathway Method refinement
In focusAll three placeholders — from placeholder to check rule
From three readers to check roles
The home page names three readers — the source, the person and the third reader. Why is that enough to design a check method, and what has to change before a check can run?
Three placeholders are enough to design a check method. To run checks, each one opens into roles that a check rule can name.
System of record · Golden record · System of engagement · Release decision · Instructed person (electrically) · Skilled person (electrically) · Data owner · Data lineage · Action approval gate (AI agent)
Read the cardContext cardTransitional topic
The cow model: working through a topic in breadth and depth
How do you work through a topic in breadth and depth before you publish on it?
Saina Veigel’s cow model as a recipe: frame, four passes, the reader’s test, the remainder and the habit — and what each step rests on.
Inversion (thinking tool) · Curse of knowledge · Second-order thinking · Metacognition · Illusion of explanatory depth · Dual-process theory · Analysis and synthesis (method) · Mental model (thinking tool) · Latticework of mental models · Divergent and convergent thinking · Incubation (problem solving) · Spacing effect · Neuroplasticity · Zettelkasten · Deliberate practice · Hebbian learning · Habit formation · Implementation intention · Reward system · Digital dementia · Encephalization · Working memory · Cognitive load theory
Read the cardPathway card Pathway Method refinement
In focusThe third reader — the machine
The machine is a network, not a monster
When an AI answer is wrong, why is “the machine got it wrong” no diagnosis?
“The machine” is a web of parts — index, retrieval, model, agent — with different roles and owners. Naming the part shows what to correct.
Retrieval-augmented generation (RAG) · Large language model (LLM) · AI agent · Excessive agency · Answer provenance · Data lineage
Read the cardPathway card Pathway Method refinement
In focusThe third reader — the answer function
The answer function compiles — it doesn’t read
Is an AI answer function a reader of the documentation?
An answer function selects units and composes a new text nobody released. It is a compiler — and has to say what its answer was made from.
Retrieval-augmented generation (RAG) · Answer provenance · Grounding · Translation memory · Release decision · Content provenance
Read the cardPathway card Pathway Method refinement
In focusThe person — the reader
Three operators, one warning
Why does the same warning need different information for an instructed operator, a trained operator and a skilled person?
Instructed, trained, skilled: the same warning needs a different task per qualification, and every qualification meets every situation.
Instructed person (electrically) · Skilled person (electrically) · Ordinary person (electrically) · Qualified person · Competent person · Target audience analysis · Skill-, rule- and knowledge-based behavior · Lockout/tagout (LOTO) · Operating mode
Read the cardPathway card Pathway Method refinement
In focusThe source — the writer
One writer, many writers
Who writes a technical text — and why does it matter that it isn’t one person?
A text has many writers: writer, expert, translator, reviewer, releasing person and AI functions. Naming them shows where questions and checks go.
Subject-matter expert (SME) · Release decision · Segregation of duties (SoD) · Translation memory · AI content disclosure · Content provenance
Read the cardContext card Method refinement
In focusAll three placeholders — the five activity families
The role grammar: readers, writers and compilers
How do you say who does what with a piece of technical information — when “the writer”, “the machine” and “the reader” are too coarse?
Four axes for everyone who writes, decides, compiles, uses or writes back technical information — people and systems alike.
Target audience analysis · Qualified person · Competent person · Instructed person (electrically) · Skilled person (electrically) · Segregation of duties (SoD) · Retrieval-augmented generation (RAG) · Action approval gate (AI agent) · Subject-matter expert (SME)
Read the cardPathway card Pathway Method refinement
In focusThe person and the third reader — job groups and the AI agent
From job groups to roles
Why does a grid of job groups — writers, marketers, project managers, developers, the reader, the AI agent — show who is involved but not who does what?
Job groups show who is involved; roles show who does what. “The AI agent” alone hides three roles: taking over, answering and writing back.
AI agent · Action approval gate (AI agent) · Excessive agency · Human in the loop · Role-based access control (RBAC) · Target audience analysis
Read the cardPathway card Pathway
From structured content to controlled evidence: five stages of AI-ready documentation
Is structured content enough to make medical-device documentation ready for AI?
Structured content is stage two of five: AI-ready device documentation also needs links, controlled versions and answers traced back to evidence.
Structured writing · DITA (Darwin Information Typing Architecture) · Component content management system (CCMS) · Content provenance · Retrieval-augmented generation (RAG) · Traceability · Configuration management · Labeling and instructions for use (MDR and IVDR) · Risk control measure (ISO 14971) · ISO 14971 · Grounding · Source of truth · Document control
Read the cardContext card
One technical file, two sets of requirements: what the AI Act adds to an MDR or IVDR file
Does a medical device with AI need a second technical file for the EU AI Act?
A device with AI keeps one technical file: the AI Act adds content on data, logging, oversight and accuracy to the MDR or IVDR file, not a second file.
AI in medical devices (MDR, IVDR and AI Act) · Single set of technical documentation (AI Act Art. 11(2)) · Technical documentation (MDR and IVDR) · Technical documentation (AI Act) · High-risk AI system (AI Act) · Annex I Section A and Section B (AI Act) · Medical Device Regulation (MDR) · In Vitro Diagnostic Medical Devices Regulation (IVDR) · Notified body (MDR and IVDR) · Conformity assessment (MDR) · Digital Omnibus on AI · Human oversight (AI Act) · Instructions for use (AI Act) · Post-market monitoring (AI Act) · Record-keeping and logs (AI Act) · Post-market surveillance (MDR and IVDR) · ISO 13485
Read the cardContext card
Claims are controlled information objects: from claim to evidence in a medical device file
How should claims about a medical device be managed so that each one is backed by evidence?
Claims are controlled objects with wording, places and evidence; a claim that drifts from its evidence is an evidence failure, not a marketing slip.
Claims matrix (medical devices) · Intended purpose (MDR) · Clinical benefit (MDR) · Clinical evaluation (MDR) · Labeling and instructions for use (MDR and IVDR) · Summary of safety and clinical performance (SSCP) · General safety and performance requirements (GSPR) · Traceability
Read the cardContext card
The intended purpose as the semantic anchor of the technical file
Why is the intended purpose the reference point of every document in a medical-device technical file?
The intended purpose is the anchor of the technical file: classification, risks, evidence and IFU all refer to it, and contradictions are evidence failures.
Intended purpose (MDR) · Technical documentation (MDR and IVDR) · Clinical evaluation (MDR) · Labeling and instructions for use (MDR and IVDR) · Clinical benefit (MDR) · Claims matrix (medical devices) · Device classification (MDR) · General safety and performance requirements (GSPR) · ISO 14971
Read the cardContext card
A medical-device technical file is not a collection of documents. It is an evidence system.
What makes a medical-device technical file an evidence system rather than a set of documents?
A technical file works through the links between its parts: requirement, risk, control, test and clinical data. A notified body follows them; an AI would too.
Technical documentation (MDR and IVDR) · General safety and performance requirements (GSPR) · Post-market surveillance (MDR and IVDR) · ISO 14971 · Clinical evaluation (MDR) · Notified body (MDR and IVDR) · Traceability · Intended purpose (MDR) · Claims matrix (medical devices) · Design history file (DHF) · Medical device file (ISO 13485) · Post-market clinical follow-up (PMCF) · Periodic safety update report (PSUR)
Read the cardRole card Functional role
Post-market signal owner: a functional role
Who makes sure that what is learned about a device in the field reaches its risk file, its IFU and its claims?
The functional role that turns post-market signals into named updates of the risk file, IFU and claims, alongside vigilance reporting.
Post-market surveillance (MDR and IVDR) · Post-market signal (medical devices) · Vigilance (MDR and IVDR) · Corrective and preventive action (CAPA)
Read the cardRole card Functional role
Model change approver: a functional role
Who decides whether an update of an AI model in a medical device stays inside its planned modifications — and who stops it when it doesn’t?
The functional role that checks AI model updates of a medical device against the modification plan and approves or stops them.
Predetermined change control plan (PCCP) · Model versioning · Change control · AI-enabled device software function (AI-DSF)
Read the cardRole card Functional role
Labeling and IFU owner: a functional role
Who makes sure the label and the instructions for use of a device say what the risk file, the evidence and the product version require?
The functional role that keeps labels and IFU linked to risk controls, evidence, product versions and languages, and releases them.
Labeling and instructions for use (MDR and IVDR) · Usability engineering file (IEC 62366-1) · Warning message · ISO 15223-1
Read the cardRole card Functional role
Clinical evidence owner: a functional role
Who keeps the clinical and performance evidence of a device in step with the IFU, the claims and the post-market data?
The functional role that keeps clinical and performance evidence current and named, so IFU, claims and summaries refer to the same figures.
Clinical evaluation (MDR) · Clinical benefit (MDR) · Post-market clinical follow-up (PMCF) · Performance evaluation (IVDR)
Read the cardRole card Functional role
Risk management file owner: a functional role
Who keeps the risk management file of a medical device connected to the requirements, tests and warnings that depend on it?
The functional role that keeps the ISO 14971 risk management file and its links to requirements, verification, labeling and post-market data.
ISO 14971 · Risk control measure (ISO 14971) · Benefit-risk determination (MDR) · Hazardous situation
Read the cardRole card Functional role
Intended purpose owner: a functional role
Who keeps the intended purpose of a medical device consistent across every document that uses it?
The functional role that keeps one controlled intended purpose for a medical device and starts the impact assessment when it changes.
Intended purpose (MDR) · Medical device (MDR) · Device classification (MDR) · Claims matrix (medical devices)
Read the cardPathway map Map
The medical-device evidence chain: from intended purpose to post-market evidence
How do the parts of a medical device’s technical documentation connect — and what has to be in place before an AI function may answer from it?
Pathway map of medical-device documentation: intended purpose, risk, IFU, clinical evidence, AI functions and post-market data as one evidence chain.
Medical Device Regulation (MDR) · In Vitro Diagnostic Medical Devices Regulation (IVDR) · Medical device (MDR) · Device classification (MDR) · Device classification (IVDR) · Rule 11: software classification (MDR) · ISO 13485 · Quality Management System Regulation (QMSR, 21 CFR Part 820) · Medical device file (ISO 13485) · Document control · ISO 14971 · Hazard · Hazardous situation · Harm · Risk control measure (ISO 14971) · Benefit-risk determination (MDR) · AI in medical devices (MDR, IVDR and AI Act) · High-risk AI system (AI Act) · Single set of technical documentation (AI Act Art. 11(2)) · Provider (AI Act) · FDA premarket pathways: 510(k), De Novo and PMA · Predetermined change control plan (PCCP) · Clinical decision support software (FDA) · Intended purpose (MDR) · Technical documentation (MDR and IVDR) · General safety and performance requirements (GSPR) · Traceability · Evidence chain · Design controls (medical devices) · Design history file (DHF) · Device master record (DMR) and device history record (DHR) · Labeling and instructions for use (MDR and IVDR) · Warning message · Usability engineering file (IEC 62366-1) · Hazard-related use scenario · Clinical evaluation (MDR) · Clinical benefit (MDR) · Performance evaluation (IVDR) · Post-market clinical follow-up (PMCF) · Summary of safety and clinical performance (SSCP) · Claims matrix (medical devices) · Post-market surveillance (MDR and IVDR) · Vigilance (MDR and IVDR) · Periodic safety update report (PSUR) · Corrective and preventive action (CAPA) · Post-market signal (medical devices) · Release decision · Unique Device Identification (UDI) · Evidence version · Change control · Software as a medical device (SaMD) · Medical device software (MDSW) · AI-enabled device software function (AI-DSF) · IEC 62304 · Software of unknown provenance (SOUP) · Software bill of materials (SBOM) · IEC 81001-5-1 · Cybersecurity · Training data · Representativeness of training data (medical AI) · Data lineage · Data governance · Model versioning · Human oversight (AI Act) · Instructions for use (AI Act) · Transparency obligations (AI Act) · Model card · Model monitoring · Model drift · Post-market monitoring (AI Act) · Terminology management · Taxonomy · ISO 15223-1 · Knowledge graph · Evidence architecture · Ontology · Single-sourcing · Configuration management · Version control · Translation memory · Retrieval-augmented generation (RAG) · Access control · Grounding · Source of truth · Human in the loop
Read the cardContext card
What an answer function may read: retrieval permissions for a medical-device technical file
Which parts of a medical-device technical file may an AI answer function retrieve?
An answer function may only read what is released for its reader, in the right version and language; confidential, personal and draft content stays out.
Retrieval-augmented generation (RAG) · Access control · Role-based access control (RBAC) · Technical documentation (MDR and IVDR) · Vigilance (MDR and IVDR) · Document control · Source of truth · Content provenance · Grounding · Machine translation post-editing (MTPE)
Read the cardContext card
Manufacturer, provider, deployer: who is who when a medical device contains AI
Which AI Act role does a medical device manufacturer, a hospital or an importer take on?
The MDR manufacturer is usually the AI Act provider; the hospital using the device is a deployer. Roles in both laws match in name, not always in duties.
Provider (AI Act) · Deployer (AI Act) · Economic operators (MDR and IVDR) · AI in medical devices (MDR, IVDR and AI Act) · Medical Device Regulation (MDR) · Human oversight (AI Act) · Instructions for use (AI Act) · In-house devices (MDR and IVDR Article 5(5))
Read the cardContext card
Model version, software version, device version: three numbers that are not the same
What is the difference between the model version, the software version and the device version of an AI-enabled medical device?
Model, software and device version identify different things; after a model update all three must be traceable to each other and to the evidence.
Unique Device Identification (UDI) · IEC 62304 · Configuration management · Software of unknown provenance (SOUP) · Predetermined change control plan (PCCP) · AI drift · Model card · Software release · EUDAMED
Read the cardContext card
Training data provenance is part of the device file, not a data-science record
Why does the origin of training data belong in the technical documentation of an AI-enabled medical device?
Claimed device performance is only as valid as the data behind it: training data provenance belongs in the technical file, not only in data science tools.
Representativeness of training data (medical AI) · Content provenance · Data lineage · Data governance · Training data · Bias (AI) · AI-enabled device software function (AI-DSF) · Clinical evaluation (MDR) · W3C PROV
Read the cardContext card
Documenting the limits of a model: what the IFU of an AI-enabled device must say
How should the limitations of an AI model be documented for the users of a medical device?
Model limits belong in the IFU as information for safety, tied to evidence: validated population, declared performance, failure modes, what the user checks.
Instructions for use (AI Act) · Labeling and instructions for use (MDR and IVDR) · Model card · Representativeness of training data (medical AI) · Human oversight (AI Act) · AI-enabled device software function (AI-DSF) · AI drift · Hallucination (AI)
Read the cardPathway card Pathway
From a locked model to planned change: documenting model updates of a medical device
How can an AI-enabled medical device change its model without every update becoming a new regulatory question?
A PCCP (USA) or pre-determined changes under AI Act Art. 43(4) let a device model change within documented bounds; outside them, the old rules apply.
Predetermined change control plan (PCCP) · AI-enabled device software function (AI-DSF) · AI drift · FDA premarket pathways: 510(k), De Novo and PMA · Technical documentation (AI Act) · Configuration management · Design controls (medical devices) · IEC 62304 · Technical documentation (MDR and IVDR)
Read the cardContext card
Who decides which value counts: which system decides a value of product information?
When product information passes through PLM, ERP, content systems, spreadsheets, portals and AI functions, how do you tell which system decides a value and which one only holds a copy?
Ten roles systems play with product information, one test question per value and a small lineage record, after a blueprint by Saina Veigel.
Source of truth · Master data · Master-data synchronization · Product lifecycle management (PLM) · Enterprise resource planning (ERP) · Component content management system (CCMS) · Bill of materials (BOM) · Configuration management · Configuration baseline · Digital thread · Integration layer · Integration platform · Data pipeline · Data lineage · Data governance · Data quality · Metadata · iiRDS Consumer · Semantic interoperability · Retrieval-augmented generation (RAG) · Answer provenance · Knowledge graph · Excessive agency · Action approval gate (AI agent) · Human in the loop · Audit trail · Release decision · Content provenance · Shadow AI · System of record · Golden record · Master data management (MDM) · Data owner · Data steward · Data custodian · Source system · Downstream system · Derived data · End-user computing (EUC) · Semantic model (BI) · ETL and ELT · Data product · Data contract · W3C PROV · System of engagement · Product information management (PIM)
Read the cardContext card
Emotion and pain experiments on language models: what they measure and what they leave open
When a study makes a language model “afraid”, “sad” or “in pain”, what has it shown — and what not?
Emotion and pain experiments on language models measure functional states and changed behavior. What they show, what they leave open, how to write about them.
Evaluation (evals) · Computational functionalism · Biological naturalism · Model welfare · Anthropomorphism · Moral patient · Interpretability (AI)
Read the cardContext card
Reporting duties after an AI agent incident: AI Act, CRA, NIS 2 and GDPR
When an AI agent has caused an incident, which EU reporting duties can apply — who reports, to whom, and by when?
AI Act, CRA, NIS 2 and GDPR side by side: which reporting duties can follow an AI agent incident, who reports and by when.
Serious incident (AI Act) · Severe incident (CRA) · Incident reporting (reporting obligations) · NIS 2 Directive · Post-market monitoring (AI Act) · Record-keeping and logs (AI Act) · Deployer (AI Act) · Provider (AI Act) · Cyber Resilience Act (CRA) · EU AI Act
Read the cardPathway map Map
When an AI agent crosses the line: detect, contain and restore operations
What has to be in place when an AI agent does something it isn’t allowed to do — in IT systems, IoT fleets, OT and on machines — and how is it detected, stopped and released again?
Pathway map: how to detect, contain and recover from an AI agent crossing its boundary in IT, IoT, OT and on machines — four role routes, 23 stations.
Agent permission scope · Excessive agency · AI agent · Tool allowlist (AI agent) · Agent identity · Principle of least privilege · Action approval gate (AI agent) · Irreversible action (AI agent) · Human in the loop · Agent trace · AI interaction log · OpenTelemetry · Audit trail · Canary token · Anomaly detection · Security monitoring · Rogue agent (AI) · Blast radius (AI agent) · Agent goal hijack · Security incident · Incident response · Agent containment · Agent kill switch · Credential revocation · Stop button or similar procedure (AI Act) · Record-keeping and logs (AI Act) · Serious incident (AI Act) · Severe incident (CRA) · Incident reporting (reporting obligations) · NIS 2 Directive · Return to service (after an incident) · Rollback · Compensating action · Restart prevention · Post-incident review · Root-cause analysis · Management of change (MOC) · Enterprise resource planning (ERP) · Sandboxing · Rollback capability · Continuous deployment · Configuration management · Over-the-air update (OTA) · Building automation · Energy management system (EMS) · Rate limiting · Cyber Resilience Act (CRA) · IoT gateway · Internet of Things (IoT) · Historian · Manufacturing execution system (MES) · Human-machine interface (HMI) · SCADA software · Distributed control system (DCS) · Batch control · Zone and conduit model · PLC programming · Programmable logic controller (PLC) · Industrial robot · Collaborative robot · Robot controller · Emergency stop · Safety function · Autonomous mobile robot (AMR) · Automated guided vehicle (AGV) · Fleet management system · Controlled stop · Coordinated restart · Operating mode · Machinery Regulation (EU) 2023/1230
Read the cardContext card
Permission, capability, mandate: the three parts of an AI agent’s boundary
Which boundary does an AI agent cross when it “goes too far” — and why does it matter which one?
An AI agent’s boundary has three parts — permission, capability, mandate. Each can be crossed alone and needs its own control.
Agent permission scope · Tool allowlist (AI agent) · Excessive agency · Agent goal hijack · Rogue agent (AI) · Agent identity · Blast radius (AI agent) · Principle of least privilege · AI agent
Read the cardContext card
Detecting an AI agent that crosses its boundary
How do you notice that an AI agent has done something it wasn’t allowed to do — before a customer, an operator or the month-end report does?
How to notice an AI agent crossing its boundary: traces, refused calls, unusual patterns, canary tokens — and why good results prove nothing.
Agent trace · Canary token · Anomaly detection · Security monitoring · Rogue agent (AI) · Agent goal hijack · Tool allowlist (AI agent) · AI interaction log · OpenTelemetry
Read the cardContext card
Containing an AI agent: pause, revoke, isolate, shut down
How do you stop an AI agent that has crossed its boundary — without making things worse, and without confusing it with stopping a machine?
Containing an AI agent in four rungs — pause, revoke, isolate, shut down — and why an agent stop is not a machine’s emergency stop.
Agent containment · Agent kill switch · Credential revocation · Stop button or similar procedure (AI Act) · Emergency stop · Safe state · Zone and conduit model · Network segmentation · Blast radius (AI agent) · Controlled stop
Read the cardContext card
Restoring operations after an AI agent incident
What can be undone after an AI agent crossed its boundary, what can’t — and when may the agent and the systems it touched run again?
After an AI agent incident: what rollback can undo, what needs compensating actions, and the release before agent, systems and machines run again.
Return to service (after an incident) · Post-incident review · Rollback · Rollback capability · Compensating action · Irreversible action (AI agent) · Restart prevention · Coordinated restart · Over-the-air update (OTA) · Management of change (MOC)
Read the cardContext card
Provider rules, national law, EU law and human rights: who sets the rules for AI
Which texts govern what an AI system may say and do — and whom does each of them bind?
Human rights, constitutions, EU and national law, co-regulation and provider rules for AI: who sets each layer, whom it binds, who enforces it.
Constitution (law) · Charter of Fundamental Rights of the European Union · Primacy of EU law · Canadian Charter of Rights and Freedoms · Constitution of the People's Republic of China · International Covenant on Civil and Political Rights (ICCPR) · Council of Europe Framework Convention on AI · Co-regulation (AI) · Provider usage policy (AI) · Model specification (AI) · EU AI Act · Provider (AI Act) · Deployer (AI Act) · Fundamental rights impact assessment (AI Act) · GPAI Code of Practice · Harmonized standard · NIST AI Risk Management Framework · Executive Order 14365 · California SB 53 (Transparency in Frontier AI Act) · Artificial Intelligence and Data Act (AIDA) · Canadian Voluntary Code of Conduct on Generative AI · Directive on Automated Decision-Making (Canada) · Interim Measures for Generative AI Services (China) · AI-generated content labeling measures (China) · Legal personhood (AI) · AI usage policy · Horizontal AI regulation
Read the cardContext card
In-band and out-of-band supervision of an AI agent
What is the difference between in-band and out-of-band supervision of an AI agent?
In-band agent controls run inside the agent runtime; out-of-band controls watch from a separate context and can intervene independently.
AI agent · Runtime environment · Access control · Logging · System boundary
Read the cardPathway card Pathway
Indirect prompt injection in agent chains
How does indirect prompt injection travel through a chain of AI agents, and what changes when the chain gets control points?
How indirect prompt injection moves through AI agent chains, what control points change, what they don't, and a nine-step path to a controlled chain.
Prompt injection · Prompt · AI agent · Evaluation (evals) · Test harness · Authorization · Principle of least privilege · Audit trail · Human in the loop · Logging · Agent containment · Agent goal hijack · Excessive agency
Read the cardContext card
Unauthorized access and the duty to notify
Why are stopping unauthorized agent access and notifying the affected party two separate duties?
Blocking an unauthorized action and reporting it are different duties with different owners, evidence and deadlines.
Authorization · Access control · Incident response · Logging · Responsibility assignment
Read the cardContext card
AI systems and the machine: scope and out of scope of every level
Where do a company's AI systems end, where does the machine reach into them, and what lies outside every configuration?
Five levels, the machine across them and two boundaries: what AI systems and the AI environment cover, where the machine reaches in and who decides.
AI system (AI Act) · General-purpose AI model (AI Act) · Provider (AI Act) · Deployer (AI Act) · Foundation model · Large language model (LLM) · Retrieval-augmented generation (RAG) · Knowledge graph · AI agent · API gateway · System-of-systems · AI management system (AIMS) · AI usage policy · AI governance
Read the cardContext card
Workflow-scoped AI and open-ended agents
What is the difference between AI embedded in a defined workflow and an open-ended AI agent?
Workflow-scoped AI is bound to a procedure step, roles and released knowledge; an open-ended agent picks its own path within external limits.
AI agent · Technical documentation · Version release · Responsibility assignment · Validation
Read the cardContext card
The agent master record as a documentation object
What should an agent master record for an AI agent contain?
The agent master record links an AI agent's purpose, delegated permissions, runtime configuration, logs and owners into one traceable object.
AI agent · Technical documentation · Configuration management · Responsibility assignment · Authorization · Logging
Read the cardContext card
Claims about AI consciousness: four questions that get merged
Which different questions hide behind a sentence such as "the AI is not conscious"?
AI consciousness claims merge four questions: evidence, design, safety, and moral-legal status. With a context matrix and a dated log of the debate.
Claim (assurance) · AI system (AI Act) · Risk · Categories (Aristotle) · Transparency obligations (AI Act) · Evaluation (evals) · Human oversight (AI Act) · Moral patient · Computational functionalism · Biological naturalism · Anthropomorphism · Sycophancy (AI) · AI companion · Model welfare · Legal personhood (AI) · Model specification (AI) · Provider usage policy (AI) · Constitution (law)
Read the cardContext card
From behavior objective to evaluation criterion
How does a written behavior rule become a score in a model evaluation?
A behavior rule becomes a score via sub-behaviors, prompts and a rubric; the score shows conformity to the rule, not the truth of its premises.
Evaluation (evals) · Requirement · Verification · Claim (assurance)
Read the cardContext card
What a connectivity server does between an AI system and a machine
What does an industrial connectivity server or integration platform do between an AI system and a machine?
A connectivity server normalizes controller data to OPC UA or MQTT and turns a released AI result into one defined write or method call.
OPC Unified Architecture (OPC UA) · Integration layer · MQTT · OPC · Information model
Read the cardContext card
Model behavior policy and model card: two documents, two time directions
What is a model behavior policy, and how does it differ from a model card?
A model behavior policy states how a model should behave in future versions; a model card reports what one released version actually did.
Model card · Training vs. inference · AI governance · Technical documentation
Read the cardContext card
Agent workloads vs. microservices and batch jobs
How does an AI agent workload differ from a microservice or a batch job in operations?
Agent workloads are long-lived, stateful and non-deterministic — which is why orchestration for agents differs from microservices and batch jobs.
Orchestration · Microservice architecture · Checkpoint (software) · AI agent · Process (computing)
Read the cardContext card
Explainability and action traceability
What is the difference between explainability and action traceability for an acting AI system?
Explainability says why a result was produced; action traceability records what was triggered, allowed, executed and verified.
Explainability (AI) · Observability · Logging · Change control
Read the cardContext card
Judging a curated collection of sources
What makes a curated collection of responses or sources usable as evidence?
A curated source collection is usable as evidence when its selection rules, standpoint, known-item checks and change history are published.
Source code repository · Accountability (AI) · Bias (AI) · View (systems engineering)
Read the cardContext cardTransitional topic
Working memory and procedure steps: why "seven plus or minus two" is the wrong rule
Does working memory research justify a fixed maximum number of steps in a procedure, such as seven plus or minus two?
Miller's 1956 paper set no step limit; Cowan puts working memory at about four chunks. Steps stay on the page, so design each step, not the step count.
DITA task · Instructions for use (IFU) · Usability · Safety information usability · Mental model
Read the cardContext cardTransitional topic
Cognitive psychology in technical communication: theory, application and the bridges between them
What is the difference between cognitive psychology as theory and its application in documentation, and how do the two fit together?
Theory describes how readers perceive, remember and decide; application turns it into documentation practice. Bridges make each step visible and checkable.
Mental model · Human error · Warning message · Safety information usability · Usability · Risk assessment · Instructions for use (IFU) · DITA concept · DITA task
Read the cardContext cardTransitional topic
Human error types and troubleshooting: from slips, lapses and mistakes to recovery information
How do the error types of James Reason and Jens Rasmussen's skill-, rule- and knowledge-based behavior help design troubleshooting and recovery information?
Slips, lapses and mistakes need different help: design against slips, checklists against lapses, symptom-cause-remedy tables and system models for mistakes.
Human error · Troubleshooting · DITA troubleshooting topic · Reasonably foreseeable misuse · Risk assessment · Alarm management · Mental model
Read the cardContext cardTransitional topic
Cognitive load and split attention: the three types of load, the split-attention effect and what they mean for documentation
What is cognitive load, how does the split-attention effect relate to it, and which design decisions in technical documentation follow from the two?
Cognitive load theory names three types of load; split attention is one avoidable cause of extraneous load. What follows for figures, text and examples.
Cognitive load theory · Intrinsic cognitive load · Extraneous cognitive load · Split-attention effect · Working memory · Cognitive schema · Chunking (working memory) · Expertise reversal effect · Worked example · Redundancy principle (multimedia learning) · Spatial contiguity principle · Coherence principle (multimedia learning) · Signaling principle (multimedia learning) · Cognitive theory of multimedia learning · Progressive disclosure · Minimalism (technical communication) · Think-aloud protocol · Cognitive walkthrough · Symbol · Safety information readability · Usability · DITA task · Instructions for use (IFU) · Safety sign
Read the cardContext cardTransitional topic
Mental models and concept information: why readers need to know how it works before they act
How does research on mental models support placing concept information before task information in technical documentation?
Readers act on a mental model built from the system image, documentation included. Concept information before tasks helps them predict and recover.
Mental model · DITA concept · DITA task · Human-machine interface (HMI) · Instructions for use (IFU) · Operating mode · Mental model (thinking tool)
Read the cardContext cardTransitional topic
Attention and warning placement: from selective attention and habituation to where warnings go
What does research on attention and warnings say about where and how warnings should be placed in instructions?
A warning works only if noticed, read, understood and acted on. Place it right before the step with the hazard, and keep warnings rare enough to stay seen.
Warning message · Safety information usability · Safety information readability · Safety sign · Warning (HMI message) · Alarm flooding · Risk assessment · Residual risk
Read the cardContext cardTransitional topic
How first principles thinking evolved: an adaptation of many adaptations
Where does first principles thinking come from, and how has it changed on its way to technical writing?
First principles thinking runs from Aristotle via Euclid, Descartes, Boyd and Musk to mental models; its version for technical writers is one more adaptation.
First principles thinking · First principle (philosophy) · Socratic questioning · Axiomatic method · Analysis and synthesis (method) · Cartesian method · First-principles calculation · Destruction and creation (John Boyd) · Reasoning by analogy · Mental model · Latticework of mental models · Five whys · Risk assessment · Assumption (systems engineering)
Read the cardContext card
Documenting foreseeable misuse without making it an option
How do instructions describe reasonably foreseeable misuse without presenting it as a way to use the product?
Foreseeable misuse is stated as a clear boundary with its reason, kept apart from intended use and procedures, never written like an operating step.
Reasonably foreseeable misuse · Intended use · Residual risk · Risk assessment · Information for use · Machinery Regulation (EU) 2023/1230 · ISO 12100 · Warning message
Read the cardContext card
Operating modes in technical documentation
Why do operating modes change what the documentation of a machine has to say?
Each operating mode changes who works where, which safeguards act and which hazards remain, so each mode needs its own documentation decisions.
Operating mode · Operating-mode management · Residual risk · Reasonably foreseeable misuse · Safeguard · Enabling device · Lockout/tagout (LOTO) · Personal protective equipment (PPE) · Qualified person · Machinery Regulation (EU) 2023/1230 · ISO 12100
Read the cardContext card
Where a warning belongs: safety chapter or before the task
Should a warning go into the safety chapter or directly before the step it concerns?
General safety information goes into the safety chapter; a warning about a hazard in a specific action belongs directly before that action.
Warning message · Safety chapter · Residual risk · Risk assessment · Information for use · DITA hazard statement · Three-step method · Safety information usability
Read the cardContext card
Task analysis: one task, three lenses
Is task analysis one method, or are technical, collaborative and cognitive task analysis different things?
Hierarchical, team and cognitive task analysis look at the same task: its steps and plans, its handovers between roles, and the thinking behind it.
Usability · DITA task · Task allocation · Human error · RACI matrix · Training needs analysis
Read the cardContext card
Betriebsanweisung: one German word, many operator documents
What is a Betriebsanweisung, and how does it differ from the manufacturer’s operating manual?
A Betriebsanweisung is the employer’s binding instruction to its staff, not the manufacturer’s operating manual. Companies use the word for many documents.
Ordinance on Industrial Safety and Health (BetrSichV) · Ordinance on Hazardous Substances (GefStoffV) · Operating manual · Instructions for use (IFU) · Personal protective equipment (PPE) · Lockout/tagout (LOTO)
Read the cardContext card
Task analysis across the documentation disciplines
Who uses task analysis, and which documents does it feed?
Safety, regulatory, quality, R&D, technical writing, roll-out and operators all use task analysis — for different documents with different audiences.
Risk assessment · ISO 12100 · IEC 62366-1 · Training needs analysis · Quality documentation · Product lifecycle management (PLM)
Read the cardContext card
Instructions for use: one manual or a set of information products?
Do "instructions for use" mean one user manual or a collection of documents?
Instructions for use can be one manual or a set of information products for several audiences; the set is handed over and also kept in the technical file.
Instructions for use (IFU) · Information for use · Operating manual · Maintenance manual · Technical file · Labeling and instructions for use (MDR and IVDR)
Read the cardContext card
Internal and external technical documentation: what "external" really means
Does "external documentation" mean documentation that is published?
"External" documentation leaves the manufacturer; it is not necessarily public. Function, audience, owner and access decide what a document is.
Technical documentation · Technical file · Information for use · Instructions for use (IFU) · Operating manual · Machinery Regulation (EU) 2023/1230
Read the cardContext card
The Radio Equipment Directive: when sector law absorbs aspect requirements
Why does the cybersecurity part of the Radio Equipment Directive create parallel compliance paths?
RED is sector law that also carries aspect requirements; its cybersecurity part duplicated the CRA until the repeal effective December 11, 2027.
Cyber Resilience Act (CRA) · Low Voltage Directive · Electromagnetic compatibility (EMC) · Harmonized standard · Presumption of conformity · CE marking
Read the cardContext card
BSI guidance on the CRA: context-dependent, not contradictory
When two pieces of BSI guidance on the Cyber Resilience Act seem irreconcilable, is the guidance contradictory?
BSI TR-03183-1 ties every control to a risk scenario; apparent conflicts are resolved by the risk assessment, not irreconcilable advice.
Cyber Resilience Act (CRA) · Security risk assessment · Harmonized standard · Presumption of conformity · Software bill of materials (SBOM) · State of the art
Read the cardContext card
Module H under the CRA: BSI TR-03183-H builds full quality assurance on ISO/IEC 27001
How can a manufacturer demonstrate CRA conformity through its processes rather than product by product, and what does BSI TR-03183-H add?
BSI TR-03183-H shows how a CRA Module H quality system can rest on ISO/IEC 27001, audited by a notified body and still product-specific.
Conformity assessment · Notified body (MDR and IVDR) · Information security management system (ISMS) · Quality management · Security development lifecycle (SDL) · Harmonized standard · Presumption of conformity · Cyber Resilience Act (CRA)
Read the cardContext card
Receiving vulnerability reports under BSI TR-03183-3: report, notification, advisory
What does BSI TR-03183-3 expect a manufacturer to have in place before the first vulnerability report arrives?
BSI TR-03183-3 sets up the intake for vulnerability reports: security.txt, PSIRT and CSIRT, CVD policy, response times and disclosure.
Coordinated vulnerability disclosure (CVD) · Vulnerability disclosure · Security advisory · Common Vulnerability Scoring System (CVSS) · Vulnerability management · Cyber Resilience Act (CRA)
Read the cardContext card
The SBOM under BSI TR-03183-2: an inventory, not a vulnerability report
What does BSI TR-03183-2 require of a software bill of materials, and how does it relate to the CRA?
BSI TR-03183-2 defines SBOM format, fields and depth; the SBOM inventories components and never carries vulnerability information.
Software bill of materials (SBOM) · Cyber Resilience Act (CRA) · Supply chain security · License management · Security advisory · Common Vulnerabilities and Exposures (CVE) · Vulnerability management
Read the cardContext card
CRA product categories: default, important and critical
Which CRA products need a notified body, and how many products fall into the stricter categories?
Most CRA products are in the default category with self-assessment; Annex III and IV list the important and critical exceptions.
Cyber Resilience Act (CRA) · Conformity assessment · Notified body (MDR and IVDR) · Harmonized standard · Presumption of conformity · Declaration of Conformity · CE marking
Read the cardContext card
EN 50742: safety-related security levels and IEC 62443 working together
Is EN 50742 a sign that legislators misunderstand technicalities, and do its SRSLs contradict IEC 62443?
EN 50742 is a CENELEC standard, not law: SRSLs set a safety function's protection need, IEC 62443 supplies the controls; they combine.
Machinery Regulation (EU) 2023/1230 · IEC 62443 · Security level (SL) · Safety-security convergence · Tamper resistance · Harmonized standard · Presumption of conformity · Risk assessment
Read the cardContext card
Machinery Regulation and CRA: coupled through standards, not cross-references
Why can the Machinery Regulation and the Cyber Resilience Act not simply refer to each other?
MR sets safety objectives, the CRA cybersecurity objectives; scopes differ, so consistency comes from standards, not direct legal cross-references.
Machinery Regulation (EU) 2023/1230 · Cyber Resilience Act (CRA) · Harmonized standard · Safety-security convergence · Safety-security interface · IEC 62443 · Presumption of conformity
Read the cardContext card
Retrofit and substantial modification under the Machinery Regulation
When does a retrofit make the operator the manufacturer of a machine?
A retrofit turns the modifier into a manufacturer only if it is a substantial modification; every retrofit needs the risk assessment that decides it.
Retrofit · Retrofit under the Machinery Regulation · Substantial modification · Machinery Regulation (EU) 2023/1230 · Machinery Directive 2006/42/EC · Assembly of machinery · Stop controls of an assembly of machinery (1.2.4.4) · Ordinance on Industrial Safety and Health (BetrSichV) · Technical file · Safety PLC · Remote software update · Configuration management
Read the cardContext card
BetrSichV or GefStoffV: which ordinance carries the duty?
Does the duty to keep the explosion protection document current come from the BetrSichV or the GefStoffV?
The explosion protection document rests on § 6 GefStoffV, not on the BetrSichV; same result, different legal basis — a frequent mix-up.
Ordinance on Industrial Safety and Health (BetrSichV) · Ordinance on Hazardous Substances (GefStoffV) · Explosion protection document · Presumption effect (technical rules) · Adaptation to the state of the art (work equipment) · Grandfathering (work equipment)
Read the cardContext card
Grandfathering covers the design, never the file
Does grandfathering protect an existing plant and its documentation?
No general grandfathering for work equipment in Germany: older designs may stay if safe, outdated risk assessments and documents never do.
Grandfathering (work equipment) · Ordinance on Industrial Safety and Health (BetrSichV) · Adaptation to the state of the art (work equipment) · Presumption effect (technical rules) · As-built documentation · Explosion protection document · State of the art
Read the cardContext card
The explosion protection document after a conversion
When does the explosion protection document have to be updated?
The explosion protection document must follow every significant change of the plant; an extension or a moved booth is such a change (§ 6 GefStoffV).
Explosion protection document · Hazardous area classification (Ex zones) · Ordinance on Hazardous Substances (GefStoffV) · Grandfathering (work equipment) · Management of change (MOC) · Explosion protection · Retrofit
Read the cardContext card
Topic-based vs. task-based documentation
Is topic-based documentation the same as task-based documentation?
Topic-based documentation defines the unit of content; task-based documentation defines which content is needed. The two are distinct and often combined.
DITA topic · DITA task · DITA concept · DITA reference · DITA map · DITA information typing
Read the cardContext card
Minimalism and structured authoring
How do minimalism and structured authoring work together?
Minimalism defines what content does for the user; structured authoring defines its form. Content models can turn minimalist principles into rules.
Schema (data) · Schema validation · DITA (Darwin Information Typing Architecture) · DITA task · DITA specialization · Information for use
Read the cardContext card
Every Page is Page One vs. semantic documentation
Is Every Page is Page One a form of semantic documentation?
EPPO is a writing model for self-contained pages; semantic documentation makes meaning machine-readable. Distinct models that strengthen each other.
DITA topic · Metadata · Taxonomy · Ontology · Information model · iiRDS Topic · Knowledge graph
Read the cardContext card
Six technical documentation models compared
How do topic-based, task-based, EPPO, semantic, minimalist and structured documentation differ?
Six documentation models answer six questions — unit, selection, independence, meaning, action and form — and are usually combined, not chosen.
Technical documentation · DITA (Darwin Information Typing Architecture) · DITA topic · DITA task · Metadata · Information model · Schema (data)
Read the cardContext card
SDK vs. API
What is the difference between an SDK and an API?
An API is the contract between two pieces of software; an SDK is the kit of libraries, tools and samples that makes using that contract easier.
Software development kit (SDK) · Application programming interface (API) · REST API · Software library · Authentication · Interface contract · Semantic versioning
Read the cardContext card
How the four SDK types interoperate
Where do developer, publishing, platform and hardware SDKs meet in practice?
Hardware, developer, platform and publishing SDKs meet at gateways, APIs, extensions and documentation pipelines — held together by contracts and formats.
Interoperability · Interface contract · Serialization format · JSON · Distributed tracing · Observability · IoT gateway · Digital twin · Software bill of materials (SBOM)
Read the cardContext card
Layers of a hardware SDK
Which layers does a hardware SDK provide between application code and a device?
A hardware SDK stacks toolchain, drivers, abstraction layer, runtime and libraries between application code and a device, plus tools to debug it.
Device driver · Firmware · Microcontroller (MCU) · Real-time operating system (RTOS) · Compiler · Debugging · Abstraction · Application binary interface (ABI) · Embedded software
Read the cardContext card
Extension points of a platform
How does a platform SDK let outside code extend a host application?
A platform SDK opens a host application at named extension points; a manifest declares contributions and permissions, and the host keeps control.
Software framework · Application programming interface (API) · Authorization · Interface contract · Backward compatibility · Deprecation path · Low-code/no-code platform
Read the cardContext card
Four types of SDK compared
What distinguishes developer, publishing, platform and hardware SDKs?
Developer, publishing, platform and hardware SDKs share one idea — a kit for building on something — but differ in target, artifacts, users and control.
Software development kit (SDK) · Application programming interface (API) · DITA (Darwin Information Typing Architecture) · Device driver · Firmware · Software library · Compiler
Read the cardContext card
Stages of a publishing pipeline
What happens to structured content inside a publishing SDK?
A publishing SDK resolves, validates, filters, transforms and packages structured content, so one source yields HTML, PDF and other deliverables.
DITA (Darwin Information Typing Architecture) · DITA map · DITA topic · DITA conditional processing · DITA content reference (conref) · DITA key · Schema validation · Metadata
Read the cardContext card
MQTT vs. OPC UA: transport and meaning
What is the difference between MQTT and OPC UA?
MQTT moves messages through a broker; OPC UA adds an information model, services and security. How the two differ and why they are combined.
MQTT · OPC Unified Architecture (OPC UA) · Publish-subscribe pattern · Message broker · Information model · OPC UA PubSub
Read the cardContext card
Control points for AI agents
Who controls what an AI agent is allowed to do?
An AI agent acts through identities, permissions, gateways and logs. The control points that bound it and what documentation has to record about them.
AI agent · Model Context Protocol (MCP) · Role-based access control (RBAC) · Principle of least privilege · Audit trail · Human in the loop · Logging
Read the cardContext card
AI vs. classical automation
How does AI differ from classical automation?
Classical automation executes predefined rules; AI derives its behavior from data. Why the two are combined and how AI output is bounded in operations.
Artificial intelligence (AI) · Automation · Machine learning (ML) · Programmable logic controller (PLC) · Robotic process automation (RPA) · Human in the loop
Read the card
Method refinement
Method refinement: where the blueprints sharpen their own models
The blueprints on this site speak of the writer, the machine and the reader in the singular — on purpose, because a lens in the singular brings one question into focus. These pathway cards show where that singular becomes too coarse for a decision, and how the role grammar sharpens it. The earlier blueprints stay valid; the cards show how the models develop.
- 1In focusThe source — the writerOne writer, many writersWho writes a technical text — and why does it matter that it isn’t one person?
- 2In focusThe person — the readerThree operators, one warningWhy does the same warning need different information for an instructed operator, a trained operator and a skilled person?
- 3In focusThe person and the third reader — job groups and the AI agentFrom job groups to rolesWhy does a grid of job groups — writers, marketers, project managers, developers, the reader, the AI agent — show who is involved but not who does what?
- 4In focusThe third reader — the machineThe machine is a network, not a monsterWhen an AI answer is wrong, why is “the machine got it wrong” no diagnosis?
- 5In focusThe third reader — the answer functionThe answer function compiles — it doesn’t readIs an AI answer function a reader of the documentation?
- 6In focusAll three placeholders — the five activity familiesThe role grammar: readers, writers and compilersHow do you say who does what with a piece of technical information — when “the writer”, “the machine” and “the reader” are too coarse?
- 7In focusAll three placeholders — from placeholder to check ruleFrom three readers to check rolesThe home page names three readers — the source, the person and the third reader. Why is that enough to design a check method, and what has to change before a check can run?