Book Blueprint: Landscape

This document records the editorial brief of Landscape: A Fact-Based Picture of Architecture and AI at Work, for Leaders, Architects and Citizens, a companion to The Transformed Architecture by Gunnar Nordqvist. It is a reference for authors and editors, not a manuscript chapter. The chapter briefs below are generated from scripts/outline_data.py by scripts/generate_outlines.py; edit the briefs there, not here.

Premise

Hans Rosling's Factfulness did three things at once: it replaced a broken mental picture of the world (two boxes, "developed" and "developing") with a better one (four income levels on a continuous scale, drawn the same way every time); it named the instincts that make people see the world as worse, simpler or more fixed than it is, and showed with a short test that educated people score below chance; and it gave a rule of thumb for each instinct. This book does the same for the landscape that is emerging in large European enterprises and public authorities when architecture becomes code and AI tools go to work: a unified way to visualise that landscape, a set of instincts that make people misread it, the evidence that corrects each one, and a rule of thumb for each.

The book is standalone and written for a general educated reader. Every concept it uses is defined in the book, in its own words, before it is used. The seven volumes of the series are published works by the same author; the book cites them as external sources and adopts their vocabulary by its exact names with attribution (docs/FOUNDATIONS.md). The book is a companion to the series, never a numbered volume of it.

The picture, the instincts, the rules, the stance

The picture. The notation Contour (docs/NOTATION.md): seven dimensions (autonomy level, reversibility tier, reach, usage mode, identity, control, evidence), three frames (boundary, time, comparison) and six standard views, with a fixed legend, drawable by hand in five minutes and rendered exactly from a machine-readable model (landscape/). Every figure of Nordvik Region and Hansa Mutual is rendered from a committed model.
The instincts. Nine, derived from the evidence the book could retrieve and store (data/), not by relabelling Rosling's ten: the box, binary, model, straight-line, fear, size, destiny, single-perspective and blame instincts. Each is shown to distort perception of this landscape with at least one piece of evidence. The candidates tested and dropped, with the evidence that decided it, are recorded in HANDOVER.md.
The rules. One rule of thumb per instinct, in one sentence, set off in its chapter, collected in Chapter 19 and printed on one page in Appendix B.
The stance. Possibilist: neither optimist nor pessimist, convinced that the facts allow a reasonable hope and demand work. Every statement about the future carries one of the labels observable, expected or speculation; the author's own projections are confined to the marked section of Chapter 20.

The test

The book opens with thirteen multiple-choice questions, three options each, every answer traceable to a dataset in data/ with a SOURCE.md. Where a published perception survey asked a comparable question, the chapter reports how respondents answered beside the measured answer; where none exists, the question is offered as a self-test and the book says so. The questions are defined in TEST_QUESTIONS in scripts/outline_data.py, presented in Chapter 1 and returned to in the Part II chapter that explains each.

Audience

Executives and board members who decide about AI and architecture; architects and engineers who have to explain the landscape to them; elected officials and public servants; journalists; citizens who read the news. Practitioners of the series will find the vocabulary they know; readers who have read none of the volumes need nothing but this book.

Boundaries

The book draws the landscape and corrects perceptions of it; it does not repeat the series' guidance on how to govern, build, defend or manage, and refers the reader to the volumes for that.
Every number is cited to a dataset or document the author can open; no statistic is invented, rounded beyond the source or extended beyond its period. Vendor surveys are marked as such in captions. Estimates say "estimate", give the method and are labelled speculation if the method is weak.
Legal content is attributed to the instrument and article, verified against the live text, and framed as organisational and architectural implication, not legal advice. The AI Act is treated as amended by the Digital Omnibus (Regulation (EU) 2026/1744).
Security and misuse content stays at the level of evidence and governance.
Effects on people are described by what studies measured; the book never calls anyone ignorant, stupid, lazy or replaceable.
No anthropomorphism: models and agents rank, draft, propose and execute; they decide only where a decision was formally delegated, and then the text says by whom.
No personal anecdote is fabricated for the author; where one would carry a chapter, a marked placeholder says what story would fit (HANDOVER.md lists them).
Factfulness is an acknowledged inspiration: none of its text, charts or questions is reproduced, its instinct names are not reused as a set, and no Gapminder or Dollar Street name, data or imagery is used.

Red threads

The notation, used the same way in every chapter, with the two organisations' landscapes rendered from the committed models.
The test, presented in Chapter 1 and returned to question by question in Part II and again in Chapter 20.
The rules of thumb, one per instinct, collected in Chapter 19 and Appendix B.
The labels observable, expected and speculation on every statement about the future.
The two organisations, Nordvik Region and Hansa Mutual, introduced in Chapter 1, drawn in every chapter and followed through 2027 in Chapter 18.

Status

Chapters are drafted in waves (Part I first, so that the notation and the test are fixed; then Part II; then Parts III and IV). The status field of each brief records outline, draft or complete.

Length and weighting

Twenty numbered chapters of 3,000 to 4,500 words of prose each, shorter and more figure-dense than the series: at least two figures per chapter and about 60 to 80 figures in all, most of them data charts or notation views. The total should land near 65,000 to 85,000 words.

Chapter briefs

Part I: The Test and the Picture

Chapter 1: Introduction: A Test, a Picture and a Stance

File: 01_introduction.md
Status: draft
Purpose: Open the book as Rosling opened his: with a test the reader takes before reading on, the measured answers, and the admission that most of us, the author included, carry a picture of this landscape that is years out of date. Present the fourteen questions, say where each answer comes from and, where a published survey asked people a comparable question, how respondents answered beside the measured answer; where none exists, say that the question is a self-test. State the author's stance in the first person: a possibilist, neither optimist nor pessimist, convinced that the facts allow a reasonable hope and demand work; acknowledge Factfulness as the model and say exactly what is borrowed (the method: a picture, instincts, evidence, rules) and what is not (none of its text, charts, questions or instinct names). State the two honest differences from Rosling's data: the data on this landscape are younger, thinner and more often produced by interested parties, and the book says so figure by figure. Introduce the two organisations, Nordvik Region and Hansa Mutual, as illustrative and never as data. Explain how to read the book, and the three labels observable, expected and speculation.
Builds on: Q1, Q8, Q9, N1, N5
Test questions: Question 1, Question 2, Question 3, Question 4, Question 5, Question 6, Question 7, Question 8, Question 9, Question 10, Question 11, Question 12, Question 13, Question 14
Evidence: Every question's dataset in data/ (Eurostat isoc_eb_ai, isoc_cicce_use, isoc_ai_iaiu, isoc_cisce_ic, isoc_sks_itspt; Digital Decade indicators; the AI Incident Database; the AI Act timeline; the Stack Overflow survey); chart_test_answers: the fourteen measured answers as small horizontal bars with the three options marked
Notation views: V1
Sections: The test; The answers; Why we get them wrong; A picture drawn the same way every time; Younger, thinner data; The stance; Two organisations; How to read this book
Practical guides: Taking the test with a team and comparing answers before anyone sees the measured ones
Worked examples: The test as a one-page sheet (the questions and options only) that can be printed
Nordvik Region and Hansa Mutual: A regional director at Nordvik Region and an executive board member at Hansa Mutual take the test in the same week and both place their own organisation's AI use higher and the EU average higher than the data show; the first one-page landscape of each organisation (V1, 1 January 2027) is shown without commentary, to be read in Chapter 4.
Observable, expected, speculation: The adoption and use figures are observable; the claim that educated readers score below chance on this landscape is the author's expectation until a survey measures it, and the chapter says so.
Figures: chart_test_answers; view_nordvik_landscape_2027-01-01; view_hansa_landscape_2027-01-01

Chapter 2: The Old Pictures: Boxes, Binaries and the Cloud

File: 02_the_old_pictures.md
Status: draft
Purpose: Show the pictures people actually carry of this landscape and why each fails against the data: the box-and-line architecture diagram (components and arrows, no people, no decision, no time), 'the AI system' as one box, humans versus machines as two camps, and 'the cloud' as one cloud. Use the survey of 284 diagrams of the seven volumes in docs/FOUNDATIONS.md as evidence that even careful practitioners draw the same things differently every time, and the Eurostat technology and cloud breakdowns as evidence that the one-box picture hides most of what matters. Make the case, in the manner of Rosling's chapter on the two boxes, that a better picture is possible because the data lie on scales, not in boxes.
Builds on: N1, N2, N3, N4, N5, N6, Q2, Q3, F11, F37
Test questions: Question 4, Question 5, Question 10
Evidence: Eurostat isoc_eb_ai: AI technology types used by enterprises (text mining, natural language generation, machine learning, process automation, image recognition, speech recognition, autonomous movement, generated media), EU 2025, by size class; Eurostat isoc_cicce_use: cloud services bought by type, EU 2014 to 2025; The diagram survey table of docs/FOUNDATIONS.md: 284 diagrams, 275 flowcharts, no legend, shape without meaning; chart_ai_technology_types: the eight AI technology types, EU 2025, all enterprises and large enterprises; chart_cloud_service_types: the cloud services bought by type over time; chart_diagram_survey: what the seven volumes drew and how (counts from FOUNDATIONS.md)
Notation views: V1
Sections: Boxes and arrows; The AI system; Humans versus machines; The cloud; Even the careful draw it differently; What a better picture needs
Practical guides: Testing a diagram: five questions that show what it hides (who acts, with whose identity, at which level, who decides, how it changes)
Worked examples: The same AI use drawn three ways, as the series drew it, side by side with the mark of the notation
Nordvik Region and Hansa Mutual: The AI estates of Nordvik Region and Hansa Mutual as their own earlier volumes drew them (as boxes in subgraphs) set against the first one-page landscape of each; the reader sees what the boxes hid: the empty corner, the seats, the ambient AI nobody chose.
Observable, expected, speculation: Observable throughout; the chapter makes no forecast.
Figures: chart_ai_technology_types; chart_cloud_service_types; chart_diagram_survey; view_nordvik_landscape_2027-01-01

Chapter 3: The New Landscape: Who Acts, With Whose Identity, Under Whose Decision

File: 03_the_new_landscape.md
Status: draft
Purpose: Define the new landscape in the book's own words, from scratch and before any notation: what it means that architecture is code (versioned, tested, delivered through pipelines; decisions as records), what it means that AI works alongside people (assistants acting with a person's identity) and as agents (software acting with its own identity, at a level of autonomy, inside a boundary, under a policy point, producing evidence), and how all of it changes over time. Adopt the series' vocabulary by its exact names with one-sentence attribution at first use: the six usage modes, the five autonomy levels, the policy decision point outside the model, containment, evidence as against attestation, decision records, mandates. Ground every concept in a scene: an email quarantined by a vendor's model nobody chose, a caseworker's draft, an agent routing a claim. Anchor the definitions in the measured landscape with the Eurostat breakdowns and the Stack Overflow data on how developers use AI tools.
Builds on: F1, F11, F12, F13, F27, F37, F38, F39, F50, F62, F72, Q3, Q7
Test questions: Question 6, Question 8, Question 13
Evidence: Eurostat isoc_ai_iaiu: individuals' use of generative AI for work, private purposes and education, by age, EU 2025; Stack Overflow Developer Survey 2024 and 2025: AI tools in the development process; trust; agents in the workflow (where retrieved); Eurostat isoc_sks_itspt: ICT specialists as a share of employment, 2004 to 2025; chart_genai_use_by_age: generative AI use by purpose and age group, EU 2025; chart_developers_ai_use_trust: use versus trust among developers (vendor-independent survey, marked as a self-selected sample)
Notation views: V1; V4
Sections: Architecture became code; AI beside people; AI as agents; Six ways an organisation comes to use a model; Boundaries, policy points and containment; Evidence; Who decides; A landscape in motion
Practical guides: Walking an organisation's AI estate with the seven questions the dimensions ask
Worked examples: A one-paragraph description of each unit of Nordvik Region's landscape in the book's vocabulary, as the model records it
Nordvik Region and Hansa Mutual: Each definition is illustrated by one unit of the two models: the ambient AI in licensed software (Mode 0, Level 4, no seat) at both; the staff assistant at Nordvik (a person's identity, Level 1); the claims-triage agent at Hansa (its own identity, Level 3, a filled diamond); the citizen-service assistant at Nordvik (Level 4 at R2, a square in the top row short of the empty corner).
Observable, expected, speculation: The vocabulary and the use figures are observable; the chapter labels as expected the spread of agents from Hansa-like organisations to Nordvik-like ones.
Figures: chart_genai_use_by_age; chart_developers_ai_use_trust; view_hansa_landscape_2027-01-01; view_hansa_decisions_2027-01-01

Chapter 4: The Notation: One Picture, Drawn the Same Way Every Time

File: 04_the_notation.md
Status: draft
Purpose: Teach the notation to the reader, scene by scene: name it, show the one-page landscape of Nordvik Region and read it aloud dimension by dimension (columns, rows, size, colour, shape, diamond and seat, baseline), then show Hansa Mutual's and compare. Explain each of the seven dimensions with the reason it was chosen (the conclusion of the series behind it, cited), define the four scales the book introduces (reversibility tiers, reach classes, decision tiers) and the two it adopts (modes, levels), show the legend, state what is deliberately left out and why, and give the thirty-second reading order. Say that the notation is drawable by hand in five minutes and that the renderer makes the sketch exact, not possible, and point to Appendix A (the reference card) and Appendix B (the hand-drawing pages).
Builds on: N1, N4, N5, N6, N7, Q3, Q6, F38, F39, F72, F84
Evidence: No data chart: the chapter's figures are notation views rendered from the two models
Notation views: V1; V3; V4; V5; V6
Sections: A first landscape, read aloud; Columns: what the AI may do; Rows: what it would take to undo; Size, colour, shape; Diamond, seat, baseline; The frames; The six views; What is left out; Pen and paper
Practical guides: Reading a landscape in thirty seconds (the six steps)
Worked examples: The legend as a margin strip; the model fragment behind one mark
Nordvik Region and Hansa Mutual: The chapter is built on the two organisations' landscapes: V1 for each, V3 for Hansa Mutual (seven jurisdictions), V4 for Nordvik Region (the Level 3 open-shift class inside a Level 2 unit), V5 for both, and V6 comparing them on 1 January 2027.
Observable, expected, speculation: Observable: the notation describes what is. No forecast.
Figures: view_nordvik_landscape_2027-01-01; view_hansa_landscape_2027-01-01; view_hansa_supply_2027-01-01; view_nordvik_decisions_2027-01-01; view_nordvik_evidence_2027-01-01; view_compare_hansa_nordvik_2027-01-01

Part II: The Instincts

Chapter 5: The Box Instinct: AI Is Not One Thing

File: 05_the_box_instinct.md
Status: draft
Purpose: Show that 'AI' in an organisation is not one thing and 'the cloud' is not one cloud. Open with a documented public scene (a board paper, a parliamentary question or a news report that speaks of 'the AI' as one system) or an illustrative one at Nordvik Region. Return to Questions 4, 5 and 10. Show with Eurostat that enterprises using AI use several technologies for several purposes, that the mix differs by size class, and that cloud use is a stack of services bought at different rates since 2014. Correct the picture with the one-page landscape, where the one box becomes many marks of different colours, shapes and sizes, and with the supply view, where the one cloud becomes containers by role and jurisdiction. Give the rule and the practice.
Builds on: N5, N3, Q2, F11, F72
Instinct: The box instinct: Drawing AI as one box, and 'the cloud' as one cloud, when the measured landscape is many uses of many kinds at many levels.
Test questions: Question 4, Question 5, Question 10
Evidence: Eurostat isoc_eb_ai: at least one, two and three technologies; the technology types, by size class, 2021 to 2025; Eurostat isoc_eb_ai_purpose: purposes of AI use, 2025; Eurostat isoc_cicce_use: services by type and sophistication level, 2014 to 2025; chart_ai_technologies_per_enterprise: one, two, three or more technologies among AI users; chart_ai_purposes: the seven purposes among AI users, all and large enterprises; chart_cloud_stack: cloud services by type, 2014 to 2025
Notation views: V1; V3
Rule of thumb (provisional): When someone says "the AI" or "the cloud", ask which ones, and count them before you judge them.
Sections: One box; The questions; The instinct; The evidence; The picture corrected; The rule of thumb; What to do
Practical guides: Turning 'we use AI' into a count: the AI inventory as the first draft of a landscape model
Worked examples: A landscape model fragment listing the thirteen units of Nordvik Region with their modes
Nordvik Region and Hansa Mutual: Nordvik Region's regional executive committee receives a paper that speaks of 'the AI system'; the one-page landscape shows thirteen units in five modes; Hansa Mutual's board paper counts seventeen. Illustrative.
Observable, expected, speculation: Observable; the growth of the number of units per organisation is expected.
Figures: chart_ai_technologies_per_enterprise; chart_ai_purposes; chart_cloud_stack; view_nordvik_landscape_2027-01-01; view_nordvik_supply_2027-01-01

Chapter 6: The Binary Instinct: Neither Automated Nor Manual

File: 06_the_binary_instinct.md
Status: draft
Purpose: Show that automated or manual, human or machine, is a false pair: the data show levels. Open with a documented public scene of the binary (a headline on jobs 'replaced', a procurement that asks whether a system is 'automated'). Return to Questions 6 and 13. Show with the individuals' data and the developer survey that most AI at work drafts for a person who decides, that use and trust diverge, and that the five autonomy levels distribute very differently from the two boxes. Correct the picture with the one-page landscape read along the columns, and the decision-rights view where one unit holds different levels for different actions. Give the rule and the practice.
Builds on: F38, F39, Q6, N4, N6
Instinct: The binary instinct: Seeing automated or manual, human or machine, when the data show five levels of autonomy and most AI at work drafts for a person who decides.
Test questions: Question 6, Question 13, Question 14
Evidence: Eurostat isoc_ai_iaiu: generative AI use by purpose, 2025; Stack Overflow 2024 and 2025: use, trust, agents in the workflow; JRC AIM-WORK 2025: algorithmic management items (tasks, rosters, instructions, ratings automatically allocated) and Eurobarometer 554 QB7; The two models: distribution of units by autonomy level; chart_algorithmic_management_eu: the share of workers reporting each automated management practice, EU27, JRC AIM-WORK 2025; chart_use_versus_trust: use and trust among developers; chart_units_by_level: units per autonomy level in the two illustrative landscapes (illustrative)
Notation views: V1; V4
Rule of thumb (provisional): Ask at which level, for which action: "automated" is five answers, not one.
Sections: Automated or manual; The questions; The instinct; The evidence; The picture corrected; The rule of thumb; What to do
Practical guides: Setting the level per action class and recording it
Worked examples: A decision-rights fragment for the SOC investigation agent of Hansa Mutual (enrich at Level 4, isolate at Level 3, wipe at Level 0)
Nordvik Region and Hansa Mutual: Hansa Mutual's SOC investigation agent holds four levels for four actions; Nordvik Region's rostering system holds Level 2 for rosters and Level 3 for open shifts. Illustrative.
Observable, expected, speculation: Observable; the movement of units from Level 1 to Level 3 within bounds is expected.
Figures: chart_use_versus_trust; chart_algorithmic_management_eu; chart_units_by_level; view_hansa_decisions_2027-01-01

Chapter 7: The Model Instinct: The Use Sets the Risk and the Value

File: 07_the_model_instinct.md
Status: draft
Purpose: Show that the model is neither the risk nor the value: the use sets both. Open with a documented public scene (a debate about which model a government or company 'chose', or an incident where the deployment, not the model, failed). Return to Question 11. Show with the AI Incident Database classification and the OECD monitor (where retrievable) that most recorded harms arise in deployment contexts and uses, and with the AI Act's structure (provider and deployer; risk by use, Annex III) that the law reasons the same way. Correct the picture with the one-page landscape, which draws no model, and the supply view, where the model is a chip at the edge. Give the rule and the practice.
Builds on: F12, F13, F27, F50, Q9
Instinct: The model instinct: Treating the model as the risk and the value, when the usage mode and the use set both.
Test questions: Question 11, Question 12
Evidence: AI Incident Database: incidents by harm type, sector and deployer versus developer (where retrieved); OECD AI Incidents Monitor: counts and method (where retrieved); AI Act timeline: provider and deployer, risk by use; chart_incidents_by_harm_type: the distribution of recorded harms; chart_incidents_deployer_developer: how often the deployer and the developer differ
Notation views: V1; V3
Rule of thumb (provisional): Ask what it is used for and who deploys it before you ask which model it is.
Sections: Which model?; The questions; The instinct; The evidence; The picture corrected; The rule of thumb; What to do
Practical guides: Classifying a use before choosing a model: the questions the AI Act asks
Worked examples: Two units of Hansa Mutual on the same model family in different modes and rows
Nordvik Region and Hansa Mutual: At Hansa Mutual the same general-purpose model family sits behind the claims assistant (Level 1, R2) and, through the hybrid set-up, behind engineers' agentic tools (Level 2); at Nordvik the open-weight model behind the staff assistant is the same as behind the citizen-service assistant, two rows apart. Illustrative.
Observable, expected, speculation: Observable; the law's application dates labelled as stated in the live text.
Figures: chart_incidents_by_harm_type; chart_incidents_deployer_developer; view_hansa_supply_2027-01-01

Chapter 8: The Straight-Line Instinct: Curves That Bend

File: 08_the_straight_line_instinct.md
Status: draft
Purpose: Show that adoption and capability curves bend, stall and jump, and that extrapolating the last two points misleads in both directions. Open with the Digital Decade: a target of 75 per cent of enterprises using AI by 2030, a 2023 baseline trajectory that reached 20 per cent, and a 2026 baseline, after two fast years, that reaches 54 per cent. Return to Questions 3 and 9. Show the EU series 2021 to 2025 by size class and country, the cloud series since 2014 as a curve that did bend, and where retrieved the AI Index data on compute and benchmark saturation. Correct the picture with the landscape over time: units move cell by cell, not along a line. Give the rule and the practice.
Builds on: F25, F58, Q4, N2
Instinct: The straight-line instinct: Extending the last two points of an adoption or capability curve into the future, when the measured curves bend, stall and jump.
Test questions: Question 3, Question 9
Evidence: Digital Decade indicators: historical data, ideal trajectory and baseline trajectories computed in 2023, 2024, 2025 and 2026; Eurostat isoc_eb_ai 2021 to 2025 by size class and country; Eurostat isoc_cicce_use 2014 to 2025; Stanford AI Index: compute and benchmarks (where retrieved); chart_digital_decade_trajectories: the target, the ideal path and four baselines; chart_adoption_by_size_class: 2021 to 2025; chart_cloud_curve: a decade of cloud adoption
Notation views: V2
Rule of thumb (provisional): Never extend the last two points; ask what bent the last curve and what could bend this one.
Sections: A target and four trajectories; The questions; The instinct; The evidence; The picture corrected; The rule of thumb; What to do
Practical guides: Reading a projection: who computed it, from which points, and when it was last recomputed
Worked examples: A chart specification that draws every trajectory with its computation year
Nordvik Region and Hansa Mutual: Nordvik Region's landscape over 2027: two units move, one is retired, one appears; nothing moves along a line. Hansa Mutual's board asks for a projection of agents and receives a V2 instead. Illustrative.
Observable, expected, speculation: Observable for the series; the chapter labels every projection it shows as the publisher's expectation and makes none of its own.
Figures: chart_digital_decade_trajectories; chart_adoption_by_size_class; chart_cloud_curve; view_nordvik_year_2027-01-01_2027-12-31

Chapter 9: The Fear Instinct: The Loud Harms and the Quiet Ones

File: 09_the_fear_instinct.md
Status: draft
Purpose: Show that the dystopia crowds out the quiet, measured harms. Open with a documented public scene: what Europeans say they fear about AI (Eurobarometer, Ipsos, Pew where retrieved) set against what the incident and threat data record. Return to Questions 7 and 11. Show with Eurostat that one in five enterprises had an ICT security incident in 2024 and that most were failures, not attacks; with ENISA what the threat landscape actually consists of; with the AI Incident Database which harms recur. Correct the picture with the one-page landscape: the empty corner is empty, and the marks that need attention are the Mode 0 squares without seats and the dashed baselines at R3. Keep the series' discipline: effects on people described by what studies measured. Give the rule and the practice.
Builds on: F49, F51, F53, F58, Q8
Instinct: The fear instinct: Letting the dystopia crowd out the quiet, measured harms that fill the incident and threat data.
Test questions: Question 7, Question 11
Evidence: Eurostat isoc_cisce_ic: incidents by consequence and cause, 2022 and 2024; ENISA Threat Landscape: threat types and sectors (where retrieved); AI Incident Database by harm type; Pew 2025 (public against experts; 24 countries), Eurobarometer 554 (jobs, monitoring, automated dismissal), Ipsos AI Monitor 2025 (nervous, excited): what people fear; chart_concern_public_experts: more concerned than excited, US public against AI experts (Pew 2025), and the nine EU countries of Pew's 2025 global survey; chart_incident_causes: unavailability by failure versus attack; chart_fear_versus_record: what respondents fear beside what the record shows (where survey data allow)
Notation views: V1
Rule of thumb (provisional): When a story frightens you, find the base rate and the quiet harm it hides.
Sections: What we fear; The questions; The instinct; The evidence; The picture corrected; The rule of thumb; What to do
Practical guides: Reading an incident report: harm, cause, deployer, reversibility
Worked examples: An incident record fragment that places an incident on the landscape
Nordvik Region and Hansa Mutual: Nordvik Region's politicians ask about rogue agents; the Region has none, and its ambient AI has no seat. Hansa Mutual's board fears the IT-operations agent; the stop drill is the measured answer. Illustrative.
Observable, expected, speculation: Observable; the chapter labels the growth of agent-related incidents as expected and the catastrophic scenarios as speculation, citing AI Threat for the evidence ladder.
Figures: chart_incident_causes; chart_concern_public_experts; chart_fear_versus_record; view_hansa_landscape_2027-01-01

Chapter 10: The Size Instinct: Numbers Without Denominators

File: 10_the_size_instinct.md
Status: draft
Purpose: Show that a large number without its denominator misleads. Open with documented public numbers: billions of attacks, millions of jobs, thousands of incidents, 10 million ICT specialists. Return to Questions 1, 2 and 8. Show with Eurostat how the same adoption fact reads as 20 per cent of enterprises, 55 per cent of large ones and a different number of people; how 10.45 million ICT specialists is 5 per cent of employment; and how incident counts need deployments as denominators. Correct the picture with the notation's size scale: reach as people touched, in four classes, never transactions. Give the rule and the practice.
Builds on: F11, Q8, N6
Instinct: The size instinct: Being impressed by a large number without its denominator, its base rate or its comparison.
Test questions: Question 1, Question 2, Question 8
Evidence: Eurostat isoc_eb_ai by size class; Eurostat isoc_sks_itspt: specialists as count and as share; AI Incident Database counts per year set against adoption; chart_specialists_count_and_share: thousands and per cent, 2004 to 2025; chart_adoption_denominators: the same fact as share of all enterprises, of large enterprises and of employment
Notation views: V1
Rule of thumb (provisional): Never accept a number alone; ask what it is divided by and compared with.
Sections: Billions; The questions; The instinct; The evidence; The picture corrected; The rule of thumb; What to do
Practical guides: Three denominators to ask for: people, deployments, time
Worked examples: The reach class of each unit of Hansa Mutual with the count behind it
Nordvik Region and Hansa Mutual: Hansa Mutual's board hears that the claims-triage agent handled hundreds of thousands of claims; the landscape sizes it by the people it touched. Nordvik's 40,000 applications a month become one XL mark. Illustrative.
Observable, expected, speculation: Observable.
Figures: chart_specialists_count_and_share; chart_adoption_denominators; view_hansa_landscape_2027-01-01

Chapter 11: The Destiny Instinct: Nothing Here Is Fixed, and Nothing Is Inevitable

File: 11_the_destiny_instinct.md
Status: draft
Purpose: Show that nothing in this landscape is fixed and nothing is inevitable: architecture that was documents is now code that changes weekly, laws change (the AI Act's dates moved in 2026), trajectories are recomputed, and every delegation can be reversed if someone recorded it. Open with a documented public scene: the Digital Omnibus moving the Annex III dates, or a 'the AI revolution is inevitable' headline set against the 2021 to 2023 stall. Return to Questions 9 and 12. Show the DORA and CNCF data on change frequency and delivery (where retrieved, marked vendor and foundation surveys) and the AI Act timeline as retrieved from the live text. Correct the picture with the landscape over time and the trails. Give the rule and the practice.
Builds on: F1, F2, F25, F80, Q4, N2
Instinct: The destiny instinct: Treating architecture as fixed documents and AI as an inevitable tide, when both change by decisions that are recorded and reversible.
Test questions: Question 9, Question 12
Evidence: AI Act timeline: original and amended application dates with articles; DORA State of DevOps: deployment frequency and AI adoption effects (where retrieved; vendor survey); CNCF survey: GitOps and policy as code (where retrieved); Digital Decade trajectories recomputed each year; chart_ai_act_timeline: the dates before and after the Omnibus; chart_delivery_frequency: how often organisations change production (vendor survey, marked)
Notation views: V2
Rule of thumb (provisional): Nothing here is fixed and nothing is inevitable; ask what was decided, by whom, and what it would take to reverse it.
Sections: Inevitable; The questions; The instinct; The evidence; The picture corrected; The rule of thumb; What to do
Practical guides: Keeping a reversibility test for every delegation (the series' delegation register, in one page)
Worked examples: A decision record that retires a unit, and the trail it leaves
Nordvik Region and Hansa Mutual: Nordvik Region retires its paused screening pilot in June 2027; Hansa Mutual's board moves the decision right over production write access to itself in March. Illustrative.
Observable, expected, speculation: Observable for dates and series; expected for continued amendment; speculation marked where the chapter mentions it.
Figures: chart_ai_act_timeline; chart_delivery_frequency; view_hansa_year_2027-01-01_2027-12-31

File: 12_the_single_perspective_instinct.md
Status: draft
Purpose: Show that security, compliance and productivity are one landscape seen from three chairs, and that each chair alone misreads it. Open with a documented public scene: the same deployment described by a CISO, a compliance officer and a product leader (from published surveys of each group where retrieved: Microsoft Work Trend Index for leaders, Stack Overflow for developers, ENISA for security). Return to Question 13. Show the gaps between what leaders expect, what workers report and what security records. Correct the picture with the single landscape read three ways: the same marks, three reading orders. Give the rule and the practice.
Builds on: F62, F63, F66, Q1, Q5, N1
Instinct: The single-perspective instinct: Seeing only security, only compliance or only productivity, when the three sit on one landscape and move together.
Test questions: Question 13
Evidence: Microsoft Work Trend Index (vendor survey, marked); Stack Overflow 2025; ENISA Threat Landscape; Eurobarometer 554 (workers' views of AI at work), KPMG and University of Melbourne 2025 (employees' reported use and trust; CC BY-NC-SA, figures quoted with citation, no chart reproduced); chart_three_chairs: what leaders, workers and security respondents report, side by side, each marked by source type
Notation views: V1; V5
Rule of thumb (provisional): Before you decide, read the landscape from the other two chairs.
Sections: Three descriptions of one deployment; The questions; The instinct; The evidence; The picture corrected; The rule of thumb; What to do
Practical guides: A three-chair review of a landscape in one meeting
Worked examples: The evidence view of Nordvik Region read by an auditor, an engineer and a politician
Nordvik Region and Hansa Mutual: Nordvik Region's three readers of the same landscape; Hansa Mutual's works council as a fourth chair. Illustrative.
Observable, expected, speculation: Observable; the chapter labels convergence of the three readings as expected where regulation forces shared evidence.
Figures: chart_three_chairs; view_nordvik_evidence_2027-01-01

Chapter 13: The Blame Instinct: When Nobody Designed Who Decides

File: 13_the_blame_instinct.md
Status: draft
Purpose: Show that when something goes wrong the blame lands on the model or the vendor because nobody designed who decides. Open with a documented public incident post-mortem in which the deployer blamed the system (from the AI Incident Database or a published inquiry report retrieved in data/). Return to Question 11. Show with the incident data how often deployer and developer are distinct and how rarely a decision holder is named; with the series' conclusions why a name is necessary and not sufficient. Correct the picture with the decision-rights view: the seat above every mark, or its absence. Give the rule and the practice.
Builds on: F43, F67, F76, F78, F84, Q9, N4
Instinct: The blame instinct: Blaming the model or the vendor when something goes wrong, when the decision rights were never designed.
Test questions: Question 11
Evidence: AI Incident Database: deployer versus developer, harmed parties; Published inquiry reports where retrieved; chart_who_was_named: deployers, developers and decision holders named in incident records (where the taxonomy allows)
Notation views: V4
Rule of thumb (provisional): Look for the seat before you look for the culprit; if nobody was named to decide, that is the finding.
Sections: The post-mortem; The questions; The instinct; The evidence; The picture corrected; The rule of thumb; What to do
Practical guides: Designing decision rights before go-live: tier, holder, stop path, record
Worked examples: A decision record naming the holder and the stop path for one unit
Nordvik Region and Hansa Mutual: Both organisations' ambient AI has no seat at the start of 2027 and a seat by the end; Hansa Mutual's board takes the seat for production write access. Illustrative.
Observable, expected, speculation: Observable; the spread of named decision holders is expected where the AI Act's deployer duties apply.
Figures: chart_who_was_named; view_hansa_decisions_2027-12-31; view_nordvik_decisions_2027-12-31

Part III: Drawing Your Landscape

Chapter 14: The One-Page Landscape: Drawing It From What You Already Have

File: 14_the_one_page_landscape.md
Status: draft
Purpose: Teach the reader to draw their own one-page landscape from what the organisation already has: the AI inventory or system register, the mandate manifests, the delegation register, contracts and the proxy logs. Map each register to the model's fields explicitly (the AI system register and usage modes of AI Governance to mode, since, owner and hosting; the mandate manifests and autonomy levels of AI Red Team to level, policy point and decision holder; the delegation register of AI Boss to the decision tier of people-management functions), show how reversibility and reach are assessed, how the seat is found, and what to do when a field is unknown (draw it as unknown; the gap is the finding). Walk through the model file, the renderer and the hand-drawn sketch.
Builds on: F15, F16, F41, F57, F77, N5, N8, Q2
Evidence: No data chart; the chapter's figures are views and a model fragment
Notation views: V1
Sections: What you already have; From register to model; Assessing reversibility; Assessing reach; Finding the seat; Drawing the unknown; Rendering and sketching
Practical guides: Populating a landscape model in an afternoon
Worked examples: The full model of Nordvik Region in YAML, annotated
Nordvik Region and Hansa Mutual: Nordvik Region's AI architect populates the model from the Region's register and finds two units nobody had registered; Hansa Mutual's group platform team generates the model from its registers in a pipeline. Illustrative.
Observable, expected, speculation: Observable.
Figures: view_nordvik_landscape_2027-12-31; view_nordvik_landscape_2027-01-01

Chapter 15: The Landscape Over a Year: Reading Change

File: 15_the_landscape_over_a_year.md
Status: draft
Purpose: Teach the reader to read change: the landscape over a year, trails, new and retired marks, drift between modes, decay of reversibility, and the three labels for anything said about the future. Use the two organisations' 2027 and the measured series of Part II to show what a year of change looks like and does not look like.
Builds on: F25, F80, Q4, N2
Test questions: Question 3
Evidence: Eurostat isoc_eb_ai 2024 to 2025 by country as a 'before and after' chart; chart_country_change_2024_2025: each country's move in one year
Notation views: V2
Sections: Two dates; Drift; Decay; New and retired; Reading a year; Saying what comes next
Practical guides: A quarterly re-render and the five questions to ask of the trails
Worked examples: The timeline entries of the two models for 2027
Nordvik Region and Hansa Mutual: Both organisations' V2 for 2027 read trail by trail. Illustrative.
Observable, expected, speculation: Observable for the series; the chapter demonstrates the labels on its own statements.
Figures: view_nordvik_year_2027-01-01_2027-12-31; view_hansa_year_2027-01-01_2027-12-31; chart_country_change_2024_2025

Chapter 16: Supply, Sovereignty and Jurisdiction: Where the Landscape Runs

File: 16_supply_sovereignty_and_jurisdiction.md
Status: draft
Purpose: Teach the supply and sovereignty view: where the landscape really runs, which suppliers in which roles, which jurisdictions, and what the law says about it (the AI Act's provider and deployer roles, DORA's register of information for financial entities, the GDPR's transfer rules, framed as implications). Use the cloud data and any retrievable market-share or sovereignty data, marked by source type.
Builds on: F14, F30, F31, Q7, N3
Test questions: Question 5
Evidence: Eurostat isoc_cicce_use by sophistication level and country; EU cloud market data where retrieved (press-release data marked); chart_cloud_by_country: paid cloud use by Member State 2025; chart_cloud_market_shares: where retrievable, marked
Notation views: V3
Sections: Where it runs; Roles, not brands; Jurisdictions; What the law adds; The evidence; Reading a supply view
Practical guides: Placing every supplier in one afternoon from the contracts register
Worked examples: The suppliers block of the Hansa Mutual model
Nordvik Region and Hansa Mutual: Hansa Mutual's V3 with the UK and Swiss entities outside the EU; Nordvik Region's V3 with its ambient AI outside the EU and everything it chose inside. Illustrative.
Observable, expected, speculation: Observable; the chapter labels the European sovereign-cloud offers as expected and marks market projections as the publishers' expectations.
Figures: view_hansa_supply_2027-12-31; view_nordvik_supply_2027-12-31; chart_cloud_by_country

Chapter 17: Decision Rights and Evidence: Who Decides, and How We Know

File: 17_decision_rights_and_evidence.md
Status: draft
Purpose: Teach the decision-rights view and the evidence view together: who may decide what at which level, where the policy point enforces it, where the stop path is, and where proof comes from. Build on the series' conclusions that a name is necessary and not sufficient, that attestation is not evidence, and that Level 3 without a policy point is Level 4 with a misleading label. Show how a board, an auditor and an engineer read the two views.
Builds on: F13, F16, F39, F43, F67, F76, N4, N6, N9
Evidence: No data chart; the two models' decision and evidence views
Notation views: V4; V5
Sections: The seat; The ceiling; The diamond and the stop path; Evidence against attestation; Reading the two views
Practical guides: A decision-rights review of one unit in one page
Worked examples: A decision record and an evidence manifest fragment for one unit
Nordvik Region and Hansa Mutual: Hansa Mutual's V4 after the board took the seat for production write access; Nordvik Region's V5 before and after the supplier evidence schedule. Illustrative.
Observable, expected, speculation: Observable.
Figures: view_hansa_decisions_2027-12-31; view_nordvik_evidence_2027-12-31

Chapter 18: A Year at Nordvik Region and Hansa Mutual: Two Landscapes Compared

File: 18_a_year_at_nordvik_and_hansa.md
Status: draft
Purpose: Follow the two organisations through 2027 in the notation, quarter by quarter, and compare them: a cautious public authority that buys most of what it runs and a commercially driven insurer that builds, under one jurisdiction and under seven. Use every view, end with the comparison view at 31 December 2027, and draw the lesson that the same picture serves both and shows their differences without ranking them.
Builds on: F83, F84, N1, Q5
Evidence: No data chart; the views of both organisations
Notation views: V1; V2; V3; V4; V5; V6
Sections: January; The first half; The second half; December; Side by side
Practical guides: Running a year-end landscape review
Worked examples: The snapshots block of the two models
Nordvik Region and Hansa Mutual: The whole chapter. Illustrative throughout, consistent with the timelines in the two models and docs/EXAMPLE_ORGANISATIONS.md.
Observable, expected, speculation: Illustrative; the chapter labels its one general conclusion as the author's expectation.
Figures: view_compare_hansa_nordvik_2027-01-01; view_compare_hansa_nordvik_2027-12-31; view_nordvik_year_2027-01-01_2027-12-31; view_hansa_year_2027-01-01_2027-12-31

Part IV: Seeing Clearly

Chapter 19: The Rules of Thumb in Practice

File: 19_the_rules_of_thumb.md
Status: draft
Purpose: Collect the nine rules of thumb and show them in practice: for an executive reading a board paper, an architect reviewing a design, a public official answering a question in a council, a journalist reading a press release and a citizen reading the news. One short scene per reader, each applying several rules to a landscape. Point to Appendix B for the rules on one page.
Builds on: Q8, Q9
Evidence: No new data; the chapter reuses figures of Part II by reference
Notation views: V1; V6
Sections: The nine rules; The executive; The architect; The public official; The journalist; The citizen
Practical guides: The rules as a one-page checklist for a meeting
Worked examples: A board paper rewritten with the rules applied
Nordvik Region and Hansa Mutual: The board paper is Hansa Mutual's; the council question is Nordvik Region's. Illustrative.
Observable, expected, speculation: Observable.
Figures: view_compare_hansa_nordvik_2027-12-31

Chapter 20: The Next Years, and a Conclusion

File: 20_the_next_years_and_a_conclusion.md
Status: draft
Purpose: Say what the picture suggests about the next years, with every statement labelled observable, expected or speculation, in a section marked as the author's own projections, the only place in the book where they appear; then close briefly in the possibilist stance: the facts allow a reasonable hope and demand work. Return to the test: what the reader should now answer, and what would change the answers.
Builds on: F25, F58, F71, F84, Q4
Test questions: Question 1, Question 9, Question 12
Evidence: The series of Part II, by reference; no new data
Notation views: V1
Sections: Observable; Expected; Speculation; The author's projections; The test, again; A reasonable hope
Nordvik Region and Hansa Mutual: The two organisations' landscapes at the end of 2027 as the last figure of the book. Illustrative.
Observable, expected, speculation: The chapter is the labelled outlook of the book; the author's projections are confined to its marked section.
Figures: view_compare_hansa_nordvik_2027-12-31

Appendices

21_glossary.md: Every term the book defines, in alphabetical order: the notation's names and scales, the usage modes and autonomy levels with their origin, the data terms (denominator, base rate, trajectory, vendor survey), the three labels, and the names of the instincts.
22_appendix_a_notation_reference.md: The reference card: the seven dimensions, the scales, the legend, the six views, the file-name patterns, the schema's fields and the mapping from the series' registers, with one rendered view and the model fragment behind it.
23_appendix_b_drawing_by_hand_and_the_rules.md: Six pages, one per view, showing how to sketch it with pen and paper in five minutes, each with the renderer's exact version beside it; then the nine rules of thumb on one page.
24_appendix_c_data_sources_and_methods.md: Every dataset in data/: publisher, title, URL, retrieval date, licence, transformation and caveats; the method of the test; how perception data were matched to measured answers; how vendor surveys are marked; what was estimated and how.
25_references.md: The single bibliography: the seven volumes, Factfulness, the datasets and surveys, the legal instruments, the reports and articles.
26_about_the_author.md: The author profile, how the book was made, and the technical foundation.