Appendix C: Data Sources and Methods

This appendix is the book's account of where its numbers come from and what was done to them. Every chart in the book is drawn from a file in the repository's data/ directory, every file is traceable to a public address and a retrieval date, and every number in the prose can be found in one of those files at the precision the book prints it. The appendix explains how the sources are graded and marked, how the fourteen questions of the test were built, where a measured answer could be set beside a published survey and where it could not, under which licences the figures are reproduced, what was derived rather than read, how the figures are produced, and, dataset by dataset, who published what, when it was retrieved and what the publisher warns about. The last part is generated from the SOURCE.md file of every dataset, so that it cannot drift from the files the charts read.

How the book treats data

Rosling could draw on demographic series that statistical offices and international agencies had kept for generations, and he could put his questions to representative samples in fourteen countries and report how people answered (8, Rosling et al., 2018) (16, Gapminder Foundation, 2017). The data on this landscape differ from his in three honest ways, and the chapters say so figure by figure rather than once here.

They are younger. Eurostat's survey module on artificial intelligence in enterprises has been run in 2021, 2023, 2024 and 2025, four points, and the list of technologies it asks about gained an eighth type in 2025, so the latest aggregate rests on a longer list than the earlier ones (9, Eurostat, 2026). Eurostat's question to individuals about generative AI tools was asked for the first time in 2025, so it has one point and no series (14, Eurostat, 2026). The cloud module runs every second year and changed its list of services and its classification in 2021 (10, Eurostat, 2026). The AI Incident Database revises the counts of earlier years as reports arrive, so that the total for one year differs between snapshots (35, Responsible AI Collaborative, 2026). The law that the landscape sits under was amended in 2026 before its main obligations had applied (40, European Union, 2026).

They are thinner. Eurostat's enterprise surveys cover enterprises with ten or more persons employed in a defined set of economic activities, excluding the financial sector, and leave micro-enterprises optional; in 2025 a sample of 157,000 enterprises stood for about 1.53 million (9, Eurostat, 2026). The Eurobarometer reports that carry the only probability surveys of what Europeans think about AI at work publish their country tables as images, so this book could quote only the values that appear in their prose (29, European Commission, 2025) (30, European Commission, 2025). The perception surveys rarely ask the question the statistics answer; the next two sections say how the book handled that.

They are more often produced by interested parties. A large part of what is published about AI at work comes from vendors of AI and cloud products, from consultancies, from market-research firms and from foundations with a cause. Such figures are not false because of their origin, but they cannot be read as a statistical office's figures can, and the book grades every source it uses in five ranks:

Official statistics. Surveys run under statistical law by a statistical office, with a documented population, a published questionnaire, a known sampling and weighting method and published quality flags: Eurostat's enterprise and household surveys (9, Eurostat, 2026) (10, Eurostat, 2026) (11, Eurostat, 2026) (12, Eurostat, 2026) (13, Eurostat, 2026) (14, Eurostat, 2026) and the Digital Decade indicators that the European Commission computes from Eurostat's values (15, European Commission, 2026). These may support a statement of fact about a population. Their captions read "Source: Eurostat (online data code isoc_eb_ai), extracted 2026", and where a chart draws a country or a year that Eurostat flags, the flag follows the label: (b) for a break in series, (u) for low reliability. Eurostat's EU27 series of employed ICT specialists is flagged as estimated in every year because it aggregates national labour force surveys, and the book says so where it uses it (12, Eurostat, 2026).
Probability surveys by public or academic bodies. Surveys with a sampling frame, a stated fieldwork period, weights and a published questionnaire: Special Eurobarometer 554 and 566, face-to-face surveys of residents aged 15 and over in the 27 Member States (29, European Commission, 2025) (30, European Commission, 2025); the Joint Research Centre's AIM-WORK survey, a mobile-telephone survey of 70,316 people aged 16 to 65 in the 27 Member States, with fieldwork from 23 October 2024 to January 2025 (51, Gonzalez Vazquez et al., 2025); and the Pew Research Center's panels and national surveys (32, McClain et al., 2025) (33, Poushter et al., 2025). These may support a statement about what respondents answered, with the population, the fieldwork date and the wording of the question. Their captions name the survey, the fieldwork and "probability sample", and the sentence that uses them reads "respondents answered".
Self-selected surveys. Large, informative and representative only of the people who chose to answer: the Stack Overflow Developer Survey, with 65,437 responses in 2024 and 49,191 in 2025 (27, Stack Overflow, 2024) (28, Stack Overflow, 2025); the CNCF annual survey, drawn from the Linux Foundation's own community (21, Silverthorne & Hendrick, 2025); and the University of Melbourne and KPMG study, an online quota sample funded by its sponsor (34, Gillespie et al., 2025). These may support a statement about their respondents and never a share of a population. Their captions say "self-selected online sample" or "online panel", and the book writes "developers who answered the survey", not "developers".
Incident databases. The AI Incident Database records incidents that were reported in the press or submitted by the public and accepted by volunteer editors (35, Responsible AI Collaborative, 2026) (50, McGregor, 2021); the OECD's AI Incidents and Hazards Monitor detects events in news coverage automatically and classifies them with a model (36, OECD, 2026); ENISA's threat landscape is a curated collection of publicly reported events (37, ENISA, 2024) (38, ENISA, 2025). All three record what was reported, so their counts move with media attention, with the effort of the people who submit and with editorial throughput, as much as with the frequency of harm. They may support a statement about the record, about what kinds of harm were reported and about who was named, and never a statement about how often harm occurs. Their captions carry the snapshot or retrieval date and the words "press-reported" or "news-detected".
Vendor reports and press releases. Surveys and telemetry published by a supplier of AI or cloud services, and market estimates issued by a market-research firm as a press release: DORA's reports, published by Google Cloud (19, DeBellis et al., 2024) (20, DeBellis et al., 2025); GitHub's Octoverse (22, GitHub, 2024) (23, GitHub, 2025) (24, GitHub, 2026); Microsoft's Work Trend Index (25, Microsoft, 2025) (26, Microsoft, 2026); Ipsos's AI Monitor, a self-funded study on the firm's own online panel (31, Ipsos, 2025); and Synergy Research Group's market shares (52, Synergy Research Group, 2022) (53, Synergy Research Group, 2025). These may support a statement about what the publisher reports, and the book uses them where no better source exists for the question, marked as such. Their captions and the sentences that rest on them say "vendor survey", "vendor telemetry" or "press-release data", with the sample and the method where the publisher gives them.

Four rules follow from the grades, and they apply to every number in the book. Every number is cited, in its own sentence or the one before, to the dataset or report it comes from. No number is rounded beyond the source: where Eurostat publishes 19.95 per cent the book writes "19.95 per cent" or "about 20 per cent", never "21 per cent" (9, Eurostat, 2026). No series is extended beyond its period: where a series has four points, the chart draws four points, and a statement of fact stops at the last of them; the lines that continue past it in Chapter 8's chart of the Digital Decade trajectories are the European Commission's own projections and are labelled as such (15, European Commission, 2026). An estimate is called an estimate and its method is given, in the sentence or in this appendix, and a weak estimate is labelled speculation. Every figure of Nordvik Region and Hansa Mutual, the two fictional organisations of the book, carries the word "illustrative", because it is a drawing rendered from a model and not data.

The method of the test

The test in Chapter 1 has fourteen questions. Each was built the same way, in four steps, and the steps are recorded in data/test_answers/SOURCE.md and in the script data/_scripts/test_answers_tidy.py, which compiles the answer key from the other datasets and never from a typed number.

First, a measured answer from a dataset I can open. A question was admitted only if its answer is a value in a tidy.csv under data/, or is derived from two such values by one stated arithmetic step. Ten of the fourteen answers are published values read from the file: the share of enterprises using at least one AI technology in 2025, 19.95 per cent, and the same share among large enterprises, 55.03 per cent (9, Eurostat, 2026); the share buying paid cloud services, 52.74 per cent (10, Eurostat, 2026); the share of people aged 16 to 74 who had used a generative AI tool for work in the previous three months, 15.36 per cent (14, Eurostat, 2026); the share of enterprises with an ICT security incident in the 2024 survey, 21.54 per cent, of which 17.97 points from hardware or software failures and 3.43 from attacks (11, Eurostat, 2026); ICT specialists as 5.0 per cent of total employment (12, Eurostat, 2026); the most frequent risk domain in the AI Incident Database, with 566 of 1,498 classified incidents (35, Responsible AI Collaborative, 2026); the date from which the obligations for the high-risk AI systems of Annex III apply, 2 December 2027 (40, European Union, 2026); the share of developers who highly trust the accuracy of AI tools' output, 3.15 per cent of those who answered the question (28, Stack Overflow, 2025); and the share of EU27 workers whose tasks are allocated automatically, 21 per cent (51, Gonzalez Vazquez et al., 2025). Two answers are statements about two published values: the change from 7.65 per cent in 2021 to 8.06 per cent in 2023 (9, Eurostat, 2026) and the ranking of purposes, with marketing or sales first at 34.7 per cent of AI-using enterprises (9, Eurostat, 2026). One answer is a derived ratio: of the 19.95 per cent of enterprises that used AI in 2025, the 8.28 per cent of all enterprises that used three or more technologies make 41.5 per cent, computed as the ratio of the two published shares and labelled derived (9, Eurostat, 2026). One answer is somebody else's projection, not an observation: the European Commission's 2023 baseline trajectory for the Digital Decade target set by Decision (EU) 2022/2481 projected 20.3 per cent of enterprises using AI in 2030, against the target of 75 per cent, which the Decision words for cloud computing, big data or artificial intelligence together and the dashboard draws against AI alone, and its 2026 baseline, computed after two years of faster growth, projects 54.3 per cent (15, European Commission, 2026) (43, European Union, 2022). The chapter says that the question asks about a projection.

Second, three options with the measured answer among them. For the ten questions whose answers are numbers, the three options are round values on a scale from 0 to 100, and the two wrong options are placed at a distance chosen to test the direction of error rather than its size: one well above the measured value and one well below, or both above where the book expects the instinct to overshoot, as for the share of enterprises using AI, where the options are 20, 45 and 70 per cent and the measured value is 19.95 (9, Eurostat, 2026). For the four questions whose answers are statements or categories, the options are the measured answer and two alternatives that a reader who carries the old picture of the landscape might choose, such as "an autonomous system escaping control" beside the measured "people using AI to defraud, deceive or manipulate other people" (35, Responsible AI Collaborative, 2026). The questions and their options are fixed in scripts/outline_data.py, and every chapter refers to them by number.

Third, perception data matched only where a published survey asked a comparable question. Rosling's questions were put to representative samples, and the interesting result was how far the answers lay from the measured truth (8, Rosling et al., 2018) (16, Gapminder Foundation, 2017). This book has not put its questions to a sample, and it makes no claim about how executives, architects, officials, journalists or citizens answer them. What it does instead is to look, for every question, for a published survey that asked a comparable question of a defined population, and where one exists, to set the respondents' answer beside the measured answer with the differences in population, fieldwork date and wording stated. Comparable is a judgement, and the next section records it question by question. Where no such survey exists, the question is a self-test: the reader's own answer beside the measured one, and nothing more.

Fourth, the record. Every answer carries, in data/test_answers/tidy.csv, the dataset, the table, the indicator, the population, the geography and the year it was read from, and a method column that says whether the value is published as such or derived. The file is regenerated by the script whenever a dataset changes and is never edited by hand. Chapter 19 lists every question with the chapter that explains it.

Matching perception to measurement

The table gives, for each question, the dataset that answers it, the published survey whose respondents' answer the book sets beside it, and what differs between the two. "Self-test" means that no survey asked a comparable question of a defined population; the perception surveys the book holds ask about personal use, expectations and feelings, not about shares of enterprises, dates in a regulation or the contents of a database.

Question	Measured dataset	Perception source	What differs
Question 1: share of enterprises using at least one AI technology, 2025	Eurostat `isoc_eb_ai`, enterprises with 10 or more persons employed, EU27 (9, Eurostat, 2026)	Self-test	No survey asks respondents to estimate this share
Question 2: the same share among large enterprises	Eurostat `isoc_eb_ai`, 250 or more persons employed (9, Eurostat, 2026)	Self-test	As above
Question 3: change between 2021 and 2023	Eurostat `isoc_eb_ai`, 2021 and 2023 (9, Eurostat, 2026)	Self-test	As above; the AI module was not run in 2022
Question 4: AI users with three or more technologies	Eurostat `isoc_eb_ai`, ratio of two published shares (9, Eurostat, 2026)	Self-test	Derived value, see "What was estimated"
Question 5: enterprises buying paid cloud services, 2025	Eurostat `isoc_cicce_use` (10, Eurostat, 2026)	Self-test	No survey asks this share; the Digital Decade cloud target is measured on a narrower indicator, intermediate or sophisticated services (15, European Commission, 2026)
Question 6: people who used a generative AI tool for work, 2025	Eurostat `isoc_ai_iaiu`, individuals aged 16 to 74, previous three months (14, Eurostat, 2026)	JRC AIM-WORK: 29.9 per cent of EU27 workers used an AI-powered tool in their main job in the previous twelve months (51, Gonzalez Vazquez et al., 2025); Ipsos AI Monitor: 22 per cent of respondents in 30 countries thought it very likely and 37 per cent somewhat likely that AI would change how they do their job within five years, a vendor survey (31, Ipsos, 2025)	AIM-WORK: workers aged 16 to 65, not all individuals; any AI-powered tool used consciously, not generative AI; twelve months, not three; telephone fieldwork October 2024 to January 2025, not the 2025 household survey. Ipsos: an expectation, not use; 30 countries of which 10 Member States; online panel, 21 March to 4 April 2025
Question 7: enterprises with an ICT security incident, 2024	Eurostat `isoc_cisce_ic`, by consequence and cause (11, Eurostat, 2026)	Self-test; Chapter 9 sets the record beside what Eurobarometer 554 respondents expect of AI at work, without matching (29, European Commission, 2025)	Eurobarometer asks about jobs, management and monitoring, not about incidents
Question 8: ICT specialists as a share of employment, 2025	Eurostat `isoc_sks_itspt`, EU27 estimate from national labour force surveys (12, Eurostat, 2026)	Self-test	No survey asks this share
Question 9: the 2030 projection from the trend to 2023	Digital Decade DESI dashboard, `dd_ai` baseline trajectories of 2023 and 2026 (15, European Commission, 2026)	Self-test	The answer is the Commission's projection, not an observation
Question 10: most common purpose of AI in enterprises, 2025	Eurostat `isoc_eb_ai`, purpose indicators, share of AI-using enterprises (9, Eurostat, 2026)	Self-test	No survey asks this
Question 11: most frequent kind of harm in the AI Incident Database	AI Incident Database, snapshot of 2 November 2026, MIT risk domains of the 1,498 classified incidents (35, Responsible AI Collaborative, 2026)	Pew: 51 per cent of US adults more concerned than excited about the increased use of AI in daily life, against 15 per cent of AI experts, August to October 2024 (32, McClain et al., 2025); Eurobarometer 554: 66 per cent of EU27 respondents agree that more jobs will disappear than will be created (29, European Commission, 2025); Ipsos: 53 per cent of respondents in 30 countries agree that AI products and services make them nervous, a vendor survey (31, Ipsos, 2025)	None of the three asks which harm is most frequent; they measure the direction of concern, and the book says so. Pew: United States only; the expert sample is unweighted conference authors. Eurobarometer: residents aged 15 and over, face-to-face, April to May 2024, a question about jobs. Ipsos: online panel, 10 Member States among 30 countries
Question 12: when the Annex III high-risk obligations apply	Official Journal texts of Regulation (EU) 2024/1689 and Regulation (EU) 2026/1744 (39, European Union, 2024) (40, European Union, 2026)	Self-test	No survey asks this
Question 13: developers who highly trust the accuracy of AI tools' output	Stack Overflow Developer Survey 2025, respondents answering the trust question (28, Stack Overflow, 2025)	The survey is itself the perception; the measured side is the share of the same respondents who use the tools daily, weekly or monthly, 78.49 per cent (28, Stack Overflow, 2025)	Self-selected online sample; shares of those who answered each question, which differ in size; no probability survey asks developers this
Question 14: workers whose tasks are allocated automatically	JRC AIM-WORK, EU27 workers aged 16 to 65, fieldwork October 2024 to January 2025 (51, Gonzalez Vazquez et al., 2025)	Eurobarometer 554: 22 per cent of respondents with a current or past occupation said that digital technologies, including AI, allocate tasks to workers all the time or often, and 29 per cent that they manage worktime schedules (29, European Commission, 2025)	Population: 16 to 65 working now or in the last twelve months, against 15 and over with any current or past occupation. Mode and date: telephone, October 2024 to January 2025, against face-to-face, April to May 2024. Wording: "automatically allocated to you via a device" against "performed by digital technologies, including Artificial Intelligence". The two agree within a few points

Where the book sets two surveys side by side in one figure, as in Chapter 6's chart of algorithmic management, Chapter 9's chart of fear beside the record and Chapter 12's chart of the three chairs, the populations are not combined, the bars are not compared on one scale, and the caption names both sources and the differences.

Vendor surveys and self-selected samples

The book uses ten datasets that are neither official statistics nor probability surveys: the two Stack Overflow surveys, the CNCF survey, the University of Melbourne and KPMG study, the two DORA reports, GitHub's Octoverse posts, Microsoft's Work Trend Index reports, the Ipsos AI Monitor and Synergy's press releases. It uses them for three reasons. Some questions have no better source: no statistical office asks developers how far they trust the output of the tools they use, how often teams change production, or what share of the European cloud market European providers hold. Some of them are the picture that circulates: an executive who has read a headline share of knowledge workers using AI at work has usually read a vendor survey, and a book about how the landscape is misread must show what the circulating figures measure. And some are perception data in their own right: a self-selected survey of developers tells us what the developers who answered think, which is what Chapter 6 needs.

They are marked in three places. In the chart specification, the landscape block carries vendor_survey: true for a survey published by a supplier of AI or cloud services, and the checker refuses a spec without the field. In the caption, the words "vendor survey", "vendor telemetry", "press-release data" or "self-selected online sample" follow the source, with the sample and the method where the publisher states them: "a vendor survey of 31,000 knowledge workers in 31 markets by an online panel, leaders defined by job title" for the Work Trend Index (25, Microsoft, 2025). In the sentence, the book writes "respondents to DORA's survey" or "developers who answered", never "organisations" or "developers". The generated entries below carry the same marking.

Three further cautions apply and are repeated where the data appear. A vendor's telemetry describes the vendor's platform: GitHub's "developers" are GitHub accounts, and the Octoverse changed its scope from public repositories only to public and private between its 2025 and 2026 editions, so its editions are not one series (23, GitHub, 2025) (24, GitHub, 2026). A vendor's sample is often the vendor's audience: DORA's respondents are technology professionals recruited through DORA's channels, nearly 3,000 in 2024 and 4,867 in 2025, and the share of them who use AI at work is a share of that audience (19, DeBellis et al., 2024) (20, DeBellis et al., 2025). Microsoft's 2025 telemetry excludes European Union tenants and its 2026 survey covers only workers who already use AI, so neither describes the European workforce (25, Microsoft, 2025) (26, Microsoft, 2026). And a market share from a press release is a proprietary estimate without a public method, restated between releases: Synergy's 2022 release gave European providers 27 per cent of the European market in 2017 and 13 per cent in mid-2022, and its 2025 release gives 29 and 15 per cent for the same years (52, Synergy Research Group, 2022) (53, Synergy Research Group, 2025). The European Commission's impact assessment for the Cloud and AI Development Act relays Synergy's 15 per cent as its baseline and sets 30 per cent by 2035 as a policy objective, so the official document adds adoption, not measurement (47, European Commission, 2026) (46, European Commission, 2026).

Two sources sit between the grades and are marked accordingly. The Stanford AI Index is an academic report, but several of the series it publishes are vendor surveys it relays, among them McKinsey's adoption survey and Ipsos's public-opinion polls; where the book uses an Index series it names the originator and marks a relayed vendor survey as one (17, Maslej et al., 2025) (18, Sajadieh et al., 2026). ENISA's threat landscape is a public agency's report, but its statements on AI are a synthesis of vendor and press sources, and its incident counts are not comparable between the 2024 and 2025 editions because the method changed (37, ENISA, 2024) (38, ENISA, 2025).

Licences and what may be reproduced

Every dataset's SOURCE.md quotes the publisher's licence or terms as they read on the retrieval date, 7 November 2026, and a copy of each licence page is kept under data/_licence_evidence/. The table groups the datasets by licence class and says what the book does under each: reproduces charts it has derived from the data, with attribution; quotes figures as facts, with attribution, and reproduces nothing else; or neither. The book never reproduces a publisher's own chart, table or extended text.

Licence class	Datasets	What the book does
Eurostat free re-use of data: Commission Decision 2011/833/EU; reuse for commercial or non-commercial purposes authorised provided the source is acknowledged; modifications stated; a non-responsibility disclaimer included; data for countries outside the EU, EFTA and the candidate countries not for commercial reuse (48, Eurostat, n.d.) (45, European Commission, 2011)	`eurostat_isoc_eb_ai`, `eurostat_isoc_eb_ai_purpose`, `eurostat_isoc_cicce_use`, `eurostat_isoc_cisce_ic`, `eurostat_isoc_sks_itspt`, `eurostat_isoc_ske_itrcrs`, `eurostat_isoc_ai_iaiu`	Reproduces derived charts with the attribution "Source: Eurostat (online data code ...), extracted 2026"; states every transformation in this appendix; the Eurostat non-responsibility disclaimer stands at the end of this section. The United Kingdom rows that four of the tables carry for earlier years fall under the non-commercial exception and appear in no chart
Commission reuse policy, Decision 2011/833/EU, with CC BY 4.0 for content the EU owns; credit given and changes indicated (49, European Commission, n.d.) (45, European Commission, 2011)	`digital_decade_indicators` (which also states the European Union Public Licence in its dashboard), `eurobarometer_ai_work_2025`, `eurobarometer_digital_decade_2025`, `jrc_aimwork_2025`, `ai_act_timeline` (Official Journal texts), the Commission documents in `eu_cloud_sovereignty_or_market_share`	Reproduces derived charts with attribution to the European Commission, the dashboard or the survey, and the Eurostat acknowledgement where the values are Eurostat's; quotes and tabulates provisions of the regulations with attribution to the Official Journal; reproduces no cover photograph or illustration, which are third-party
CC BY 4.0 as stated in the report	`enisa_threat_landscape_2024`, `enisa_threat_landscape_2025` (37, ENISA, 2024) (38, ENISA, 2025)	Quotes figures with attribution to ENISA and an indication of any change; reproduces none of ENISA's charts
CC BY-SA 4.0 for the database collections; report text excluded (35, Responsible AI Collaborative, 2026)	`aiid_incidents`	Reproduces derived counts and charts with attribution to the AI Incident Database and the suggested citation (50, McGregor, 2021); keeps the derived tables in `data/aiid_incidents/tidy.csv` under CC BY-SA 4.0 because they are an adaptation of the database; reproduces no report text
CC BY-ND 4.0 (17, Maslej et al., 2025) (18, Sajadieh et al., 2026) (21, Silverthorne & Hendrick, 2025)	`stanford_ai_index_2025`, `stanford_ai_index_2026`, `cncf_annual_survey_2024`	Reproduces none of the reports' charts, cropped or whole, because the licence forbids derivatives; draws its own chart from the Index's published data files, which the Index invites readers to use, crediting the Index and the originator of each series; quotes figures from the CNCF report's text as facts with attribution
CC BY-NC-SA 4.0 (19, DeBellis et al., 2024) (20, DeBellis et al., 2025) (34, Gillespie et al., 2025)	`dora_state_of_devops_2024`, `dora_state_of_devops_2025`, `kpmg_trust_in_ai_2025`	Quotes figures only, as facts with attribution, on the author's fair-dealing judgement, because the non-commercial clause does not cover a sold book; Chapter 11's chart of delivery frequency is drawn from figures quoted in the DORA reports' text and tables, not from their artwork; reproduces no chart, table or substantial text from either publisher
ODbL 1.0 for the database, DbCL 1.0 for its contents; attribution "Stack Overflow Developer Survey, Stack Exchange Inc." (27, Stack Overflow, 2024) (28, Stack Overflow, 2025)	`stackoverflow_dev_survey_2024`, `stackoverflow_dev_survey_2025`	Reproduces derived charts of aggregates the book's own script computed from the published results file, with that attribution; the derived tables in the two `tidy.csv` files stay under the ODbL
Pew Research Center terms: reproduction, publication and sale permitted with attribution in Pew's citation format and without implying Pew's endorsement (32, McClain et al., 2025) (33, Poushter et al., 2025)	`pew_ai_public_experts_2025`, `pew_global_views_ai_2025`	Reproduces derived charts with attribution in Pew's format
Ipsos legal mentions: polls in the public domain with attribution to Ipsos, the release date, the method and a link; reproduction of the report's graphics to be cleared (31, Ipsos, 2025)	`ipsos_ai_monitor_2025`	Quotes figures only, with attribution, the release date and the method; draws no chart from the report
OECD terms and conditions; the monitor is labelled Beta and does not represent the official views of the OECD (36, OECD, 2026)	`oecd_aim_incidents`	Reproduces derived charts of the monitor's own aggregates with the attribution "OECD.AI AI Incidents and Hazards Monitor (AIM), retrieved 7 November 2026" and the Beta caveat; reproduces no article text
No Creative Commons licence; press releases and vendor reports under the publisher's terms (52, Synergy Research Group, 2022) (53, Synergy Research Group, 2025) (23, GitHub, 2025) (25, Microsoft, 2025)	`eu_cloud_sovereignty_or_market_share` (the Synergy releases), `github_octoverse_2025`, `microsoft_work_trend_index_2025`	Quotes figures only, as facts with attribution; where a figure draws a handful of such quoted figures, as Chapter 16's chart of cloud market shares and the leaders' row of Chapter 12's chart of the three chairs do, it is drawn from the quoted numbers and never from the publisher's artwork, and the caption marks it as press-release data or a vendor survey; reproduces no illustration or extended text
The author's own material	`series_diagram_survey`, `landscape_models_summary`, `test_answers`	No third-party licence applies; the counts are of the author's own files, and the two illustrative organisations are captioned "Illustrative" with no publisher named

Eurostat's terms ask for a disclaimer, and it reads as follows. The charts and figures drawn from Eurostat data in this book are the author's own derivations from the published tables; they have been computed and drawn by the author, and Eurostat bears no responsibility for them, for any transformation described in this appendix or for any conclusion the book draws from them.

What was estimated

Nothing in this book was estimated unless a chapter says "estimate". No value was interpolated, extrapolated, rounded beyond the source or extended beyond its period, and no number was typed from memory. Where a chart or a chapter uses a value that is not read as published, the value is derived by one of the arithmetic steps below, each stated in the chart's description, in the dataset's SOURCE.md or in the method column of the answer key.

Question 4, the share of AI-using enterprises with three or more technologies. The ratio of two published shares of all enterprises, 8.28 and 19.95 per cent, times 100, giving 41.5 per cent; computed by data/_scripts/test_answers_tidy.py and labelled derived (9, Eurostat, 2026).
Question 13 and the charts of developers' use and trust in Chapters 3, 6 and 12. The share who use AI tools is the sum of three published answer options, daily, weekly and monthly or infrequently, 78.49 per cent; the share who trust is the sum of highly trust and somewhat trust; both computed in the chart's transform from the published shares of respondents answering each question (28, Stack Overflow, 2025).
The Stack Overflow shares themselves. Stack Overflow publishes the raw results file, not the shares the book uses. data/stackoverflow_dev_survey_2025/compute_tidy.py and its 2024 counterpart compute each share as the count of respondents giving an answer divided by the number of respondents who answered the question, times 100, to two decimals, without weighting, for all respondents, for respondents living in the 27 Member States and for respondents in the EU27, the EFTA states and the United Kingdom; the bases are logged in compute_log.txt (27, Stack Overflow, 2024) (28, Stack Overflow, 2025).
The estimate of about 305,000 enterprises (Chapter 10). The Eurostat sample statement gives about 1.53 million enterprises with ten or more employees in the survey population, and 19.95 per cent of them used at least one AI technology in 2025; the product, about 305,000, is an estimate by multiplication under the assumption that the survey population is the population of enterprises; it carries the label estimate wherever it appears.
The chart of incidents and hazards as a share of AI news (chart_incidents_share_of_ai_news) is rendered from the OECD AIM monthly series and is not used in a chapter; it stays in charts/ as an illustration of the denominator method of Chapter 10 and is listed in charts/UNUSED.md.
Chapter 13's chart of who was named. Each of the sixty entity slugs in the database's two lists of the thirty most frequent alleged deployers and developers was classified by hand, in the chart specification, as a named organisation or an unnamed category of people or technology; the chart sums the mention counts per class and divides by the total mentions of the thirty (35, Responsible AI Collaborative, 2026).
The incident counts in Chapters 7, 9 and 13. The number of classified incidents, 1,498, is the sum of the seven published risk-domain counts; the counts by year, by risk domain, by entity and by intent are frequency counts of the snapshot's incident and classification files made by data/aiid_incidents/make_tidy.py, with one classification per incident and nothing dropped (35, Responsible AI Collaborative, 2026).
Two dates in the AI Act timeline. The entry into force of Regulation (EU) 2024/1689, 1 August 2024, is computed as the twentieth day after its publication on 12 July 2024, as its Article 113 provides, and the entry into force of Regulation (EU) 2026/1744, 27 July 2026, as the third day after its publication on 24 July 2026; both are marked as computed in the dataset, and every other date is quoted from the Official Journal text (39, European Union, 2024) (40, European Union, 2026).
The survey of the series' diagrams in Chapter 2. The counts of files, of Mermaid types, of drawing directions, of node shapes, of colour classes and of legends are exact, read from each file's syntax by data/_scripts/diagram_survey_tidy.py. The counts of diagrams touching each recurring subject come from matching keywords against node labels, edge labels and file names, and are therefore exact for what the files say and approximate for what the diagrams mean; a diagram counts under every subject it touches, so those counts add up to more than the number of files. The dataset describes the author's own seven volumes (1, Nordqvist, 2025) (2, Nordqvist, 2026) (3, Nordqvist, 2026) (4, Nordqvist, 2026) (5, Nordqvist, 2026) (6, Nordqvist, 2026) (7, Nordqvist, 2026) and supports no claim about practitioners in general.
The counts of the two illustrative landscapes in Chapter 6. data/_scripts/landscape_summary.py counts the units of Nordvik Region and Hansa Mutual per autonomy level and per usage mode at 1 January 2027 from the committed models, resolving each unit's state on that date exactly as the renderer does. The counts are illustrative, not data.

Three further things are the publishers' estimates, projections or aggregates, not the book's, and are labelled as such where they appear: Eurostat's EU27 figures for employed ICT specialists, flagged estimated in every year (12, Eurostat, 2026); the European Commission's four baseline trajectories and its ideal trajectory to the 2030 target (15, European Commission, 2026); and the Eurobarometer reports' own "total agree" and "performed all the time or often" aggregates, which the book copies as printed and does not recompute (29, European Commission, 2025). Where a chart joins two points with a dashed line, as at the 2021 break in the cloud series, the line marks a break in the published series and not an interpolated value (10, Eurostat, 2026).

How the figures are produced

The data charts are produced by a declarative pipeline, so that nothing is drawn by hand and nothing is committed that cannot be regenerated. Each chart is one Vega-Lite specification in charts/<slug>.vl.json, whose data.url names a tidy file under data/<dataset>/tidy.csv; the specification selects the rows it needs with filter, calculate and aggregate transforms and never edits the data file. The house style lives once in charts/theme.json, and every colour comes from the controlled palette docs/chart_palette.json, which python3 scripts/validate_palette.py checks for contrast and for the three dichromatic colour-vision deficiencies. python3 scripts/check_charts.py --write-missing renders every specification with vl-convert to docs/images/chart_<slug>.png and .svg and writes a manifest of the hashes of spec, theme, data and renders; python3 scripts/check_charts.py without a flag validates every specification (a landscape block with figure, source, retrieval date, licence and the vendor-survey flag; a description that names the source; no title, because the caption carries it; no pie or donut; only palette colours), renders it twice, requires the two renders to be identical and fails if a committed render differs from the fresh one. python3 -m pytest tests/test_figures.py -q checks that every image the manuscript references exists, that every chart's data file has a SOURCE.md beside it, and that every figure is followed within three lines by a caption that names a source or says the figure is illustrative. The workflow .github/workflows/verify-charts.yml runs the same checks on every change to the charts, the data or the palette and commits a regenerated render when one has changed.

python3 scripts/validate_palette.py
python3 scripts/check_charts.py --write-missing
python3 scripts/check_charts.py
python3 -m pytest tests/test_figures.py -q

The landscape views are produced by a renderer, so that every picture of an organisation is drawn from a model and never from hand-placed coordinates. The models of the two illustrative organisations are landscape/models/nordvik_region.yaml and landscape/models/hansa_mutual.yaml, validated against landscape/schema/landscape.schema.json; docs/NOTATION.md is the specification the renderer draws. python3 scripts/render_landscape.py --all renders the six standard views, V1 to V6, to docs/images/view_*.png and .svg with matplotlib's Agg backend, a fixed hash salt and no dates in the metadata, so that every run gives identical bytes; python3 scripts/render_landscape.py --check re-renders into a temporary directory and compares the bytes with the committed files, and --list prints the files the model set implies. python3 -m pytest tests/test_landscape_models.py -q validates the models against the schema, the unit identifiers and the timelines, and checks that the committed views match the models. The tidy files themselves are produced by the scripts in data/_scripts/ and in the dataset directories, run with python3 -I on the downloaded files, and python3 scripts/generate_data_appendix.py regenerates the dataset entries below from the SOURCE.md files; its --check mode fails when the entries and the files disagree.

python3 scripts/render_landscape.py --all
python3 scripts/render_landscape.py --check
python3 -m pytest tests/test_landscape_models.py -q
python3 scripts/generate_data_appendix.py --check

The datasets

The repository holds 33 datasets. Each lives in data/<slug>/ with the raw file exactly as retrieved, a SOURCE.md that records publisher, URL, retrieval date, licence, transformations and caveats, and the tidy.csv the charts read. The entries below are generated from those files by scripts/generate_data_appendix.py; caveats that several Eurostat tables share word for word (the survey population, the size classes, the reference period and the weighting) are given once, in the prose above, and not repeated here.

`ai_act_timeline`

Publications Office of the European Union (Official Journal of the European Union, L series), on behalf of the European Parliament and the Council. Application timeline of the EU AI Act, Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744 (Digital Omnibus on AI). Retrieved 2026-11-07 from http://data.europa.eu/eli/reg/2024/1689/oj. Licence: Commission Decision 2011/833/EU on the reuse of Commission documents and the Publications Office's legal notice: EU legislation published in the Official Journal may be freely reused with acknowledgement of the source ("© European Union, 1998-2026, https://eur-lex.europa.eu"); only the Official Journal in its electronic form is authentic (Regulation (EU) No 216/2013).

Caveats recorded at retrieval:

This is a timeline of application dates, not legal advice; the book should attribute each statement to the instrument and article as done here
The 2024 text stored is the original OJ publication and does not reflect corrigenda, if any, published since; the author should check EUR-Lex for corrigenda to 2024/1689 before print (EUR-Lex was not reachable from this session)
The brief's description of the Omnibus ("in force 27 July 2026, deferring Annex III high-risk obligations to 2 December 2027") is confirmed by the live text; in addition the text defers Annex I high-risk obligations to 2 August 2028, moves the national sandbox deadline from 2 August 2026 (original Art. 57(1)) to 2 August 2027, adds a 2 December 2026 date for the new Article 5 prohibitions and for the Article 50(2) transitional marking obligation, and sets 27 July 2026 for Articles 102–110

`aiid_incidents`

Responsible AI Collaborative (RAIC), AI Incident Database, https://incidentdatabase.ai. "Database Snapshots": weekly point-in-time dump backup-20261102100724.tar.bz2 (2026-11-02 10:07, 120.64 MB as listed; 120,635,874 bytes as downloaded). Retrieved 2026-11-07 from https://incidentdatabase.ai. Licence: Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0), https://creativecommons.org/licenses/by-sa/4.0/.

Caveats recorded at retrieval:

Selection and reporting bias: AIID records incidents that were reported in the press or submitted by the public and accepted by volunteer editors. Counts by year reflect media attention, submitter effort and editorial throughput as much as the underlying frequency of harm. AIID itself and the Stanford AI Index (2025, p. 167 footnote) warn that "the number of AI incidents is continually updated over time, including for previous years"; year totals in this snapshot differ from those printed in earlier reports (e.g. the AI Index 2025 reports 233 incidents for 2024; this snapshot has 313 for 2024)
Date field: date is the incident date as curated by editors; for many incidents it is the date of the first report, and some are estimated (see CSETv1 Estimated Date). Recent years are still being filled in (2026 covers only to 30 October 2026)
Taxonomy coverage: the MIT taxonomy covers 1,498 of 1,786 incidents; CSETv1 covers only 208 (mostly incidents 1–~350, classified by CSET in 2022–23) and should not be used for trends; GMF covers 41. Sector/harm distributions are therefore partial
Entities: "Alleged" deployer/developer/harmed parties are AIID's editorial attributions, not legal findings
Geography: AIID is English-language-centric and US-centric; European incidents are under-represented. The CSETv1 Location Region and Location Country fields are the only geographic fields and cover 208 incidents

`cncf_annual_survey_2024`

Cloud Native Computing Foundation (CNCF) / The Linux Foundation (Linux Foundation Research). Authors: Valerie Silverthorne and Stephen Hendrick (citation on PDF p.39). Cloud Native 2024: Approaching a Decade of Code, Cloud, and Change: the 2024 CNCF Annual Survey report, March 2025, 39 PDF pages. Retrieved 2026-11-07 from https://www.cncf.io/reports/cncf-annual-survey-2024/. Licence: "Copyright © 2025 The Linux Foundation | March 2025. This report is licensed under the Creative Commons Attribution-NoDerivatives 4.0 International Public License." (CC BY-ND 4.0, not CC BY as assumed in the brief).

`digital_decade_indicators`

European Commission, Directorate-General for Communications Networks, Content and Technology (DG CONNECT), "Digital Decade DESI" dashboard, https://digital-decade-desi.digital-strategy.ec.europa.eu/ . The underlying statistics are Eurostat's (every indicator stored here names its Eurostat source table in the data_sources column, e.g. estat_isoc_eb_ai_tany). Digital Decade DESI dashboard, chart groups "Digital Decade - EU-level trajectories" (indicator group digital_decade, "KPIs for the Digital Decade Policy Programme 2030") and "Digital Economy and Society Indicators" (indicator group desi-dtb, "Digital transformation of businesses", and desi-dsk). Retrieved 2026-11-07 from https://digital-decade-desi.digital-strategy.ec.europa.eu/. Licence: European Union Public Licence (EUPL), as stated by the dashboard for every chart group; and the European Commission reuse policy (Commission Decision 2011/833/EU), Creative Commons Attribution 4.0 International for EU-owned website content. Reproduction in this book: Reproducing derived figures in a published book is permitted under both the EUPL notice given by the dashboard and the Commission's CC BY 4.0 / Decision 2011/833/EU reuse policy, provided the source is credited, e.g. "Source: European Commission, Digital Decade DESI dashboard (indicator dd_ai), based on Eurostat (isoc_eb_ai), retrieved 2026-11-07", and changes are indicated; the EUPL is a software licence and its application to data is the dashboard's own choice, so the Eurostat acknowledgement should always accompany the figures.

Caveats recorded at retrieval:

DESI period versus reference year. A desi_2026 period means the DESI 2026 edition, which reports data collected in 2025; desi_2022 and desi_2023 both report 2021 enterprise data (the AI and cloud modules were not run in 2022), so the same value appears under two periods. Use data_year, not period, as the x-axis
dd_* versus desi_*. The dd_* rows are the Commission's EU-level KPI series for the Digital Decade Policy Programme (Decision (EU) 2022/2481) with four successive vintages of the trajectories (_2023 to _2026, named after the year of the State of the Digital Decade report that published them): dd_hist_<vintage> observed values as used in that report, dd_baseline_<vintage> projection based on past performance, dd_ideal_<vintage> the path needed to reach the target, dd_target_2030 the target. Values differ slightly between vintages because of Eurostat revisions (e.g. dd_dii 2021: 54.8 in the 2024/2025 vintages, 54.82 in the 2026 vintage). The projections are the Commission's, not observations, and must be labelled as such if charted
Cloud definition. dd_cloud and desi_cloud measure enterprises using at least one intermediate or sophisticated cloud service (Eurostat E_CC1_SI: 2021 34.01, 2023 38.97, 2025 46.69 for EU), not the broader Eurostat headline E_CC (2025: 52.74). The two must not be mixed
Big data became data analytics. desi_bigdata (Eurostat isoc_eb_bd, reference years 2016, 2018, 2020) and desi_da (Eurostat isoc_eb_das, 2023 onwards) are different questions; the dashboard's own definition says "Before 2023 this indicator was 'Big data'". dd_bigdata carries both as dd_hist_bigdata and dd_hist_da_*. There is no continuous series
Digital intensity index versions. desi_sme_di3_gelo (DII version 3, reference years 2021, 2023, 2025) and desi_sme_di4_gelo (DII version 4, 2022, 2024) count different sets of 12 technologies; the dashboard notes "The DII composition varies between different survey years". The KPI dd_dii uses DII v3 in 2021, 2023 and 2025 and the v4 value in 2022 (69.1, dd_hist_2023) and 2024 (59.312307, dd_hist_2025). Compare only within one version

`dora_state_of_devops_2024`

DORA (DevOps Research and Assessment), Google Cloud / Google LLC: vendor-published survey. Accelerate State of DevOps Report 2024 ("This is our tenth DORA report", PDF p.3); file version "v. 2024.3" (page footers); 120 PDF pages. Marked as a vendor survey in every caption. Retrieved 2026-11-07 from https://dora.dev/research/2024/dora-report/. Licence: PDF p.118: "'Accelerate State of DevOps 2024' by Google LLC is licensed under CC BY-NC-SA 4.0" (https://creativecommons.org/licenses/by-nc-sa/4.0/).

`dora_state_of_devops_2025`

DORA, Google Cloud / Google LLC: vendor-published survey. State of AI-assisted Software Development (the 2025 DORA Report), file version "v. 2025.2" (page footers), 142 PDF pages. Marked as a vendor survey in every caption. Retrieved 2026-11-07 from https://dora.dev/research/2025/dora-report/. Licence: PDF p.141: "'State of AI-assisted Software Development' by Google LLC is licensed under CC BY-NC-SA 4.0." Same implications as for the 2024 report (NonCommercial: quote figures with attribution; seek permission before reproducing DORA's own charts in a commercial book).

`enisa_threat_landscape_2024`

European Union Agency for Cybersecurity (ENISA). ENISA Threat Landscape 2024, July 2023 to June 2024, September 2024; 131 pages. ISBN 978-92-9204-675-0, DOI 10.2824/0710888 (PDF p.3). Retrieved 2026-11-07 from https://www.enisa.europa.eu/publications/enisa-threat-landscape-2024. Licence: "© European Union Agency for Cybersecurity (ENISA), 2024. This publication is licenced under CC-BY 4.0 'Unless otherwise noted, the reuse of this document is authorised under the Creative Commons Attribution 4.0 International (CC BY 4.0) licence (https://creativecommons.org/licenses/by/4.0/). This means that reuse is allowed, provided that appropriate credit is given and any changes are indicated'. For any use or reproduction of photos or other material that is not under the ENISA copyright, permission must be sought directly from the copyright holders.".

Caveats recorded at retrieval:

Open-source-based collection: "the ETL is based on publicly available information and the fact that such information might not always provide the full picture" (PDF p.10). ENISA warns (PDF p.12) that "An increase in reported cyber-attacks doesn't necessarily indicate an actual rise in the number or severity of attacks. This surge could be due to heightened media or public attention ... or threat actors claiming victims with no real impact on those victims."
Not comparable with ETL 2025 (4,875 incidents under a revised methodology)
AI statements are qualitative syntheses of open sources; the Netskope figure is a vendor figure relayed by ENISA

`enisa_threat_landscape_2025`

European Union Agency for Cybersecurity (ENISA). ENISA Threat Landscape 2025, TLP:CLEAR, October 2025; version 1.3 (file name ENISA Threat Landscape 2025_v1.3_September2026_0.pdf, the current file linked from the publication page on 2026-11-07), 89 pages. ISBN 978-92-9204-723-8, ISSN 2363-3050, DOI 10.2824/1946374 (PDF p.3). Retrieved 2026-11-07 from https://www.enisa.europa.eu/publications/enisa-threat-landscape-2025. Licence: "© European Union Agency for Cybersecurity (ENISA), 2025. This publication is licenced under CC-BY 4.0 'Unless otherwise noted, the reuse of this document is authorised under the Creative Commons Attribution 4.0 International (CC BY 4.0) licence (https://creativecommons.org/licenses/by/4.0/). This means that reuse is allowed, provided that appropriate credit is given and any changes are indicated'. For any use or reproduction of photos or other material that is not under the ENISA copyright, permission must be sought directly from the copyright holders.".

Caveats recorded at retrieval:

ENISA's dataset is "mainly based on information from open sources" plus anonymised Member State and partner information; it is a curated collection of publicly reported events, dominated by hacktivist DDoS claims that ENISA itself notes often have "little to no impact". Counts are not comparable across editions: ETL 2024 reported 11,079 incidents for July 2023–June 2024 under a different methodology (ETL 2025 uses the "ENISA Cybersecurity Threat Landscape (ENISA CTL) updated methodology published in August 2025", PDF p.9)
Statements on AI (section 4.5) are largely relayed from vendor and press sources (footnotes 72–105), including the "over 80% of phishing emails" figure from KnowBe4; treat them as ENISA's synthesis of open sources, not ENISA measurements
Version 1.3 (September 2026) is a revised file of the October 2025 report; the publication page does not list the changes

`eu_cloud_sovereignty_or_market_share`

Synergy Research Group (Reno, NV; "now part of TechInsights"). European cloud infrastructure market shares: Synergy Research Group press releases and the European Commission's Cloud and AI Development Act impact assessment. Marked as a vendor survey in every caption. Retrieved 2026-11-07 from https://www.srgresearch.com/articles/european-cloud-providers-local-market-share-now-holds-steady-at-15. Licence: © Synergy Research Group; press releases are published for citation by the press. Quoting the figures with attribution ("Synergy Research Group, press release, 24 July 2025") is the normal use; Synergy's charts are not reproduced here.

Caveats recorded at retrieval:

All market-share numbers trace back to one commercial analyst (Synergy); no official statistical source measures cloud provider market shares in Europe. Synergy revises historical shares between releases
"European market" and "European providers" are Synergy's definitions (provider headquarters), covering IaaS, PaaS and hosted private cloud; SaaS is excluded
The CADA is a proposal (adopted by the Commission 3 June 2026) and its impact assessment's 30% target is a policy objective, not a forecast; the proposal was not yet adopted by the co-legislators at retrieval time
The Eurostat 53% figure is cited here only as relayed by the Commission footnote; retrieve Eurostat's isoc_cicce_use release directly before citing it in print

`eurobarometer_ai_work_2025`

European Commission, Directorate-General for Employment, Social Affairs and Inclusion (DG EMPL), coordinated by DG Communication ("Media monitoring and Eurobarometer" unit). Fieldwork by Verian (former Kantar Public) on behalf of Verian Belgium, Eurobarometer wave 101.4. Special Eurobarometer 554. Artificial Intelligence and the future of work. Report. Catalogue number KE-01-24-018-EN-N, ISBN 978-92-68-21815-0, DOI 10.2767/8591026. © European Union, 2024. Published 13 February 2025. Retrieved 2026-11-07 from https://europa.eu/eurobarometer/surveys/detail/3222. Licence: European Commission reuse policy, Commission Decision 2011/833/EU of 12 December 2011 on the reuse of Commission documents; "Unless otherwise indicated (e.g. in individual copyright notices), content owned by the EU on this website is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) licence. This means that reuse is allowed, provided appropriate credit is given and changes are indicated." (https://commission.europa.eu/legal-notice_en, retrieved 2026-11-07). Derived figures may be reproduced in a published book with attribution to the European Commission / Eurobarometer; the cover photo (Getty Images) is third-party and excluded. Not vendor-commissioned (public authority survey; Verian is the fieldwork contractor).

Caveats recorded at retrieval:

Mirror copy, see above. The official XLSX data annex (country × answer option) was not retrievable; country-level values in tidy.csv are limited to those named in the prose
Fieldwork is April–May 2024, before the AI Act entered into force; the report was published February 2025
Trend comparisons refer to Special Eurobarometer 460 (March 2017) and 2019 waves, with modified wording ("Modified trend" notes in the questionnaire)
QB7 "performed" shares are based on the report's own description of what counts as performed (yes, all the time + yes, often); verify against the data annex before printing a split

`eurobarometer_digital_decade_2025`

European Commission, Directorate-General for Communications Networks, Content and Technology (DG CNECT), coordinated with DG Communication ("Media monitoring and Eurobarometer" unit). Fieldwork by Verian Belgium, Eurobarometer wave 103.2. Special Eurobarometer 566. The Digital Decade. Report (and Annex, Summary). Annex: catalogue number KK-01-25-042-EN-N, ISBN 978-92-68-27437-8, DOI 10.2759/0648867, © European Union, 2025. Published 16 June 2025. Retrieved 2026-11-07 from https://digital-strategy.ec.europa.eu/en/library/digital-decade-2025-special-eurobarometer. Licence: Commission Decision 2011/833/EU; CC BY 4.0 for EU-owned content, "reuse is allowed, provided appropriate credit is given and changes are indicated" (https://commission.europa.eu/legal-notice_en). Figures may be reproduced in a published book with attribution to the European Commission / Eurobarometer. Photo credits (Getty Images) are third-party. Not vendor-commissioned.

Caveats recorded at retrieval:

The full country × option tables exist only as images (annex) or on the official Eurobarometer site (https://europa.eu/eurobarometer/surveys/detail/3317 or similar; the site's API was not reachable, see RETRIEVAL_LOG.md). For a country table, OCR the annex or obtain the XLSX from the official site
This survey asks about digital policy priorities and rights, not about AI use at work; the AI-specific content is thin by design
A 2026 wave (fieldwork February–March 2026) has since been published by the Commission; this directory holds the 2025 edition as requested

`eurostat_isoc_ai_iaiu`

Eurostat (Statistical Office of the European Union), source institution "Eurostat" (JSON-stat source: ESTAT). Individuals - use of generative AI tools. Retrieved 2026-11-07 from https://doi.org/10.2908/ISOC_AI_IAIU. Licence: Eurostat free re-use of data (European Commission reuse policy, Commission Decision 2011/833/EU of 12 December 2011); editorial website content under Creative Commons Attribution 4.0 International (CC BY 4.0). Reproduction in this book: Reproducing derived figures (charts, tables, percentages) in a commercially published book is permitted provided the source is acknowledged, e.g. "Source: Eurostat (online data code isoc_ai_iaiu), extracted 2026-11-07", any modification is stated and a Eurostat non-responsibility disclaimer is included; all geographies in this table are EU Member States, EFTA members, EU candidate or potential candidate countries, or EU aggregates, so the non-commercial exception does not apply.

Caveats recorded at retrieval:

Population (ESMS 3.6): "The survey population of Individuals consists of all individuals aged 16 to 74", private households; data from the 2025 EU survey on the use of ICT in households and by individuals
Years: 2025 only; this is the first year the question was asked, so no time series exists yet
Definition (from the DESI metadata of the derived indicator desi_genai): "use of any generative AI tools (e.g. ChatGPT, Copilot, Gemini, LLaMA, Midjourney, DALL-E, Mistral) to create content like text, images, programming code, or videos in the last 3 months."
Reliability: 154 cells carry u (low reliability), 29 of them with the value suppressed, concentrated in small countries and small sub-populations (e.g. ICT professionals, students)
EU27 reference (IND_TOTAL): I_IUAI 32.66 % of all individuals (PC_IND), 34.82 % of internet users (PC_IND_IU3); I_IUAIWP 15.36 % of all individuals, 47.02 % of generative-AI users (PC_IND_IUAI)

`eurostat_isoc_cicce_use`

Eurostat (Statistical Office of the European Union), source institution "Eurostat" (JSON-stat source: ESTAT). Cloud computing services by size class of enterprise. Retrieved 2026-11-07 from https://doi.org/10.2908/ISOC_CICCE_USE. Licence: Eurostat free re-use of data (European Commission reuse policy, Commission Decision 2011/833/EU of 12 December 2011); editorial website content under Creative Commons Attribution 4.0 International (CC BY 4.0). Reproduction in this book: Reproducing derived figures (charts, tables, percentages) in a commercially published book is permitted provided the source is acknowledged, e.g. "Source: Eurostat (online data code isoc_cicce_use), extracted 2026-11-07", any modification is stated and a Eurostat non-responsibility disclaimer is included; the UK rows in this table (United Kingdom is neither an EU, EFTA nor candidate country) fall under the non-commercial exception and must be excluded from any commercial reuse.

Caveats recorded at retrieval:

Years: the cloud module was run in 2014, 2016, 2018, 2020, 2021, 2023 and 2025 for the EU aggregate; 2015 is reported by 18 geographies (BE, BG, CY, DK, EL, ES, FI, HR, HU, IE, LT, LV, MT, NO, PL, RO, SI, SK), 2017 by 20 (AT, BE, BG, CY, CZ, DK, EL, ES, FI, HR, HU, LT, LV, NO, PL, PT, RO, RS, SI, SK) and 2024 by four (ES, HR, HU, NL) only; the EU aggregates have no value in those years. Check each geography's years before charting
Break in 2021: the list of services was extended (ERP, security software, development platform added) and the sophistication classification redefined (E_CC_LO/ME/HI until 2020; E_CC1_B/I/S and E_CC1_SI from 2021). The headline E_CC is comparable across the whole period subject to the b flags (309 cells)
Which "cloud" figure? The Digital Decade 2030 target (75 %) is measured on E_CC1_SI (sophisticated or intermediate services), not on E_CC. EU27 GE10: E_CC 2021 40.97, 2023 45.32, 2025 52.74; E_CC1_SI 2021 34.01, 2023 38.97, 2025 46.69
Flag d (definition differs) appears on five CZ 2023 E_CC_DA cells
UK rows exist for 2014, 2016, 2018 and 2020 and are excluded from commercial reuse under the Eurostat exception quoted above; EU28, EU27_2007 and EU15 are historical EU aggregates

`eurostat_isoc_cisce_ic`

Eurostat (Statistical Office of the European Union), source institution "Eurostat" (JSON-stat source: ESTAT). Security incidents and consequences by size class of enterprise. Retrieved 2026-11-07 from https://doi.org/10.2908/ISOC_CISCE_IC. Licence: Eurostat free re-use of data (European Commission reuse policy, Commission Decision 2011/833/EU of 12 December 2011); editorial website content under Creative Commons Attribution 4.0 International (CC BY 4.0). Reproduction in this book: Reproducing derived figures (charts, tables, percentages) in a commercially published book is permitted provided the source is acknowledged, e.g. "Source: Eurostat (online data code isoc_cisce_ic), extracted 2026-11-07", any modification is stated and a Eurostat non-responsibility disclaimer is included; the UK rows in this table (United Kingdom is neither an EU, EFTA nor candidate country) fall under the non-commercial exception and must be excluded from any commercial reuse.

Caveats recorded at retrieval:

Years: the security module was run in 2019, 2022 and 2024 (the brief's "latest year" is 2024). The 2019 wave has fewer indicators and 175 cells are flagged b (break), mostly for 2022 versus 2019; compare 2022 and 2024 rather than 2019 to 2024 without checking flags
Reference year of the incidents: ESMS 5 says data refer to the survey period unless the model questionnaire specifies the previous calendar year; the security-incident questions in Eurostat's model questionnaires ask about incidents experienced during the previous calendar year, so verify against the model questionnaire of the survey year (linked from the ESMS page) before writing "incidents in 2023" for the 2024 column
EU27 reference (GE10, E_SEC2IANY): 2022 22.16; 2024 21.54
UK and EU28 rows exist for 2019 only; UK rows are excluded from commercial reuse under the Eurostat exception

`eurostat_isoc_eb_ai`

Eurostat (Statistical Office of the European Union), source institution "Eurostat" (JSON-stat source: ESTAT). Artificial intelligence by size class of enterprise. Retrieved 2026-11-07 from https://doi.org/10.2908/ISOC_EB_AI. Licence: Eurostat free re-use of data (European Commission reuse policy, Commission Decision 2011/833/EU of 12 December 2011); editorial website content under Creative Commons Attribution 4.0 International (CC BY 4.0). Reproduction in this book: Reproducing derived figures (charts, tables, percentages) in a commercially published book is permitted provided the source is acknowledged, e.g. "Source: Eurostat (online data code isoc_eb_ai), extracted 2026-11-07", any modification is stated and a Eurostat non-responsibility disclaimer is included; all geographies in this table are EU Member States, EFTA members, EU candidate or potential candidate countries, or EU aggregates, so the non-commercial exception does not apply.

Caveats recorded at retrieval:

Years: the AI module was run in 2021, 2023, 2024 and 2025 (not in 2022). E_AI_EC starts in 2023; E_AI_TPVSG (AI generating pictures, videos, sound/audio) has values only for 2025. The current label of E_AI_TANY reads "Enterprises using at least one of the AI technologies: AI_TTM, AI_TSR, AI_TNLG, AI_TIR, AI_TML, AI_TPA, AI_TAR, E_AI_TPVSG", i.e. it lists eight technology types; since the eighth was only asked in 2025, the 2025 aggregate rests on a longer list than 2021 to 2024. Check Eurostat's variable-specific notes for 2025 before attributing the 2024 to 2025 increase entirely to wider adoption
Breaks in series: 135 cells carry flag b (break in time series): all of them are FR and SE in 2023 and NL in 2025 (no b flags in 2021 or 2024); check the flag column before drawing a country time series
Suppressed cells: 87 confidential cells (|C) and 38 unreliable cells (u, no value) have an empty value
Definition: since 2021 the question lists the technology types individually; E_AI_TANY is a derived aggregate ("at least one of ..."), not a separately asked question
EU27 headline for reference (GE10, PC_ENT, E_AI_TANY): 2021 7.65; 2023 8.06; 2024 13.48; 2025 19.95 (exactly as in tidy.csv)

`eurostat_isoc_eb_ai_purpose`

Eurostat (Statistical Office of the European Union), source institution "Eurostat" (JSON-stat source: ESTAT). Artificial intelligence by size class of enterprise (purpose-of-use indicators). Retrieved 2026-11-07 from https://doi.org/10.2908/ISOC_EB_AI. Licence: Eurostat free re-use of data (European Commission reuse policy, Commission Decision 2011/833/EU of 12 December 2011); editorial website content under Creative Commons Attribution 4.0 International (CC BY 4.0). Reproduction in this book: Reproducing derived figures (charts, tables, percentages) in a commercially published book is permitted provided the source is acknowledged, e.g. "Source: Eurostat (online data code isoc_eb_ai), extracted 2026-11-07", any modification is stated and a Eurostat non-responsibility disclaimer is included; all geographies in this table are EU Member States, EFTA members, EU candidate or potential candidate countries, or EU aggregates, so the non-commercial exception does not apply.

Caveats recorded at retrieval:

Purpose list changed between 2021 and 2023: in 2021 the purposes were marketing/sales, production, business administration, management, logistics, ICT security and HR; from 2023 business administration and management are merged (E_AI_PBAM), HR was dropped, and accounting/finance (E_AI_PFIN) and R&D/innovation (E_AI_PRDI) were added. 2021 and 2023+ purpose shares are therefore not a continuous series except for PMS, PPP, PLOG and PITS
Denominator: PC_ENT_AI_TANY percentages are shares of AI-using enterprises; because the AI-using base is small in many countries, many of these cells are flagged u (288 cells) or suppressed as confidential (270 cells)
Years: 2021, 2023, 2024, 2025
EU27 reference values (GE10, PC_ENT_AI_TANY, 2025): marketing or sales 34.7; production processes 20.76; business administration or management 31.04; logistics 6.08; ICT security 19.77; accounting/finance 23.11; R&D/innovation 19.05; at least one purpose 73.39

`eurostat_isoc_ske_itrcrs`

Eurostat (Statistical Office of the European Union), source institution "Eurostat" (JSON-stat source: ESTAT). Enterprises that recruited or tried to recruit ICT specialists by size class of enterprise. Retrieved 2026-11-07 from https://doi.org/10.2908/ISOC_SKE_ITRCRS. Licence: Eurostat free re-use of data (European Commission reuse policy, Commission Decision 2011/833/EU of 12 December 2011); editorial website content under Creative Commons Attribution 4.0 International (CC BY 4.0). Reproduction in this book: Reproducing derived figures (charts, tables, percentages) in a commercially published book is permitted provided the source is acknowledged, e.g. "Source: Eurostat (online data code isoc_ske_itrcrs), extracted 2026-11-07", any modification is stated and a Eurostat non-responsibility disclaimer is included; the UK rows in this table (United Kingdom is neither an EU, EFTA nor candidate country) fall under the non-commercial exception and must be excluded from any commercial reuse.

Caveats recorded at retrieval:

Years: 2012, 2014, 2015, 2016, 2017, 2018, 2019, 2020, 2022, 2024 (the module is now biennial). The indicator labels say "reduced comparability with 2007"; 2007 data are not in this table
Reference year of recruitment: ESMS 5 states that questions relating to ICT specialists refer "to the calendar year prior to the survey period", so the 2024 column describes recruitment attempts and hard-to-fill vacancies during 2023
Suppressed cells: 235 u cells without value and 6 confidential cells
EU27 reference (GE10, E_ITSPVAC2, percentage of all enterprises): 2019 5.19; 2020 4.68; 2022 5.96; 2024 5.49
UK rows exist for 2012 to 2020 and are excluded from commercial reuse under the Eurostat exception

`eurostat_isoc_sks_itspt`

Eurostat (Statistical Office of the European Union), source institution "Eurostat" (JSON-stat source: ESTAT). Employed information and communications technology (ICT) specialists. Retrieved 2026-11-07 from https://doi.org/10.2908/ISOC_SKS_ITSPT. Licence: Eurostat free re-use of data (European Commission reuse policy, Commission Decision 2011/833/EU of 12 December 2011); editorial website content under Creative Commons Attribution 4.0 International (CC BY 4.0). Reproduction in this book: Reproducing derived figures (charts, tables, percentages) in a commercially published book is permitted provided the source is acknowledged, e.g. "Source: Eurostat (online data code isoc_sks_itspt), extracted 2026-11-07", any modification is stated and a Eurostat non-responsibility disclaimer is included; the UK rows in this table (United Kingdom is neither an EU, EFTA nor candidate country) fall under the non-commercial exception and must be excluded from any commercial reuse.

Caveats recorded at retrieval:

Source and population (ESMS 3.5, 3.6, 18.1): derived from the EU Labour Force Survey (EU-LFS); "Persons living in private households"; ICT specialists are defined by ISCO-08 occupation codes (ICT service managers, ICT professionals, ICT technicians, ICT installers and servicers), i.e. a "broad" occupational definition, not a sectoral one
Break in 2011 (ESMS 15.2): "Comparability over time is severely impacted by the introduction of the latest version of ISCO in 2011, which has produced a break in series" (ISCO-88 2004 to 2010, ISCO-08 from 2011). The brief asks for 2015 onwards, which lies entirely within the ISCO-08 period; the full 2004 to 2025 series is stored but should not be charted across 2010/2011
Break in 2021: the new EU-LFS methodology from 2021 (Regulation (EU) 2019/1700) produces b flags in 2021 for most geographies; the EU27 2021 value is flagged be (break, estimated)
Estimates: EU27 values are flagged e (estimated) in every year (they aggregate national LFS data); 646 cells carry e
UK rows exist for 2011 to 2019 (ESMS 3.7: "the United Kingdom (until 2019 reference year)"); they are excluded from commercial reuse under the Eurostat exception

`github_octoverse_2025`

GitHub, Inc. (Microsoft): vendor report based on GitHub's own platform telemetry; "GitHub Staff" byline. GitHub Octoverse 2025 (with the 2024 and 2026 editions). Marked as a vendor survey in every caption. Retrieved 2026-11-07 from https://github.blog/news-insights/octoverse/octoverse-a-new-developer-joins-github-every-second-as-ai-leads-typescript-to-1/. Licence: The GitHub Blog carries no Creative Commons licence; content is © GitHub, Inc. under GitHub's Terms of Service.

Caveats recorded at retrieval:

Vendor telemetry about one platform: "developers" means GitHub accounts (including bots filtered by GitHub's methods, students and inactive accounts), not employed software developers. Growth in accounts is explicitly linked by GitHub to the launch of GitHub Copilot Free; GitHub itself says of 2024 "we can't fully explain the surge in global growth"
Scope changed across editions (public-only vs public+private; Octoverse year boundaries differ); do not chain the editions as one series without noting this
Copilot-usage statistics describe GitHub's own product and are promotional in context
Posts are "Updated" after publication; the stored HTML is the version live on 2026-11-07

`ipsos_ai_monitor_2025`

Ipsos (market research company). Vendor survey: self-funded thought-leadership study run on the Ipsos Global Advisor online panel; not commissioned by a public authority. Treat as a commercial research publication. The Ipsos AI Monitor 2025. A 30-country Ipsos Global Advisor Survey, June 2025 (57-page PDF). Marked as a vendor survey in every caption. Retrieved 2026-11-07 from https://www.ipsos.com/en-id/press-release-ipsos-ai-monitor-2025. Licence: "Polls appearing on this website are deemed to be in the public domain and therefore may be used in whole or in part as long as attribution is made to the appropriate Ipsos entity that conducted the work. Outlets or organisations, wherever possible and appropriate, should note the release date, sponsor of the poll, methodology and a link to the Ipsos website where results can be obtained. Ipsos shall not be responsible for any third party's use or interpretation of the findings; however, if such use and interpretation is deemed to be inappropriate, improper, offensive, a misrepresentation or reproduced for the purpose of commercial means, Ipsos may take appropriate legal action..." The same page's copyright clause reserves site content and says reproduction requires prior written permission. In practice: derived poll figures may be reproduced with attribution to Ipsos, the release date (June 2025), the methodology and a link; reproducing the report's graphics or large verbatim parts in a commercial book should be cleared with Ipsos.

Caveats recorded at retrieval:

Vendor survey, online panel; non-probability samples in several countries (see above). "Global/30-country average" is an unweighted average of country results
The job questions (pages 46–48) are asked of all respondents except "My job", which is asked only of those with a job; the base line on the slide states 23,216 for all
Comparison values for 2023 and 2024 come from the 2025 report, not from the original 2023/2024 publications; country sets differed slightly (Switzerland not surveyed in 2023)
Country naming follows Ipsos ("Great Britain" not UK)

`jrc_aimwork_2025`

European Commission, Joint Research Centre (JRC), with DG Employment, Social Affairs and Inclusion. Fieldwork by Ipsos European Public Affairs (contractor). Gonzalez Vazquez, I., Fernandez Macias, E., Wright, S. and Villani, D., Digital Monitoring, Algorithmic Management and the Platformisation of Work in Europe, Publications Office of the European Union, Luxembourg, 2025, https://data.europa.eu/doi/10.2760/9406086, JRC143072. ISBN 978-92-68-31994-9, catalogue number KJ-01-25-490-EN-N. Companion: De Cuyper, K., Gysen, S., Vinciguerra, M.C. and Wright, S., Methodology of the AIM-WORK survey, JRC Working Papers Series on Labour, Education and Technology 2025/06, JRC143933 (not retrievable, see RETRIEVAL_LOG.md). Retrieved 2026-11-07 from https://data.europa.eu/doi/10.2760/9406086. Licence: "The reuse policy of the European Commission documents is implemented by the Commission Decision 2011/833/EU of 12 December 2011 on the reuse of Commission documents (OJ L 330, 14.12.2011, p. 39). Unless otherwise noted, the reuse of this document is authorised under the Creative Commons Attribution 4.0 International (CC BY 4.0) licence (https://creativecommons.org/licenses/by/4.0/). This means that reuse is allowed provided appropriate credit is given and any changes are indicated." Cover illustration and interior photographs are third-party (Prodigioso Volcán, Shutterstock, Adobe Stock) and excluded. Figures may be reproduced in a published book with the citation above. Public-authority study, not vendor-commissioned.

Caveats recorded at retrieval:

The AI-use measure covers AI tools "used consciously by workers", mostly chatbots; it excludes AI embedded in systems workers are unaware of (report Box, PDF p. 50)
Telephone (mobile RDD) survey of people aged 16–65; occupation rows are ISCO-08 major groups
The OECD "AI surveys of employers and workers" (2022, 7 countries) were the alternative named in the brief; the OECD site returned HTTP 403 for every request from this environment, so nothing from it is stored (see RETRIEVAL_LOG.md)

`kpmg_trust_in_ai_2025`

Gillespie, N., Lockey, S., Ward, T., Macdade, A., & Hassed, G. (2025). Trust, attitudes and use of artificial intelligence: A global study 2025. The University of Melbourne and KPMG. DOI 10.26188/28822919. Trust, attitudes and use of artificial intelligence: A global study 2025 (University of Melbourne and KPMG). Retrieved 2026-11-07 from https://assets.kpmg.com/content/dam/kpmgsites/xx/pdf/2025/05/trust-attitudes-and-use-of-ai-global-report.pdf. Licence: "Trust, attitudes and use of artificial intelligence: A global study is provided under a Creative Commons Attribution, Non-Commercial, Share Alike 4.0 International licence. You are free to use, share, reproduce and distribute the work under this licence for non-commercial purposes only, as long as you give appropriate credit to the original author(s) and the source via the citation. If any changes are made to the material, information, graphics, etc, contained in this report, the changes must be clearly indicated. Under this licence, you may not use the material for any commercial purposes. Any re-sharing of this material can only be done under the CC NC SA licence conditions." (https://creativecommons.org/licenses/by-nc-sa/4.0/).

Caveats recorded at retrieval:

Sponsor-funded; online, quota-based samples; "advanced vs emerging economy" framing is the authors'. Several EU-country samples under-represent older people (see above)
Figures with unlabeled bars (emotions by country, trustworthiness by country, confidence in entities, organisational support) are not in tidy.csv

`landscape_models_summary`

This repository. The data are illustrative: they are counts derived from the committed landscape models of the two fictional organisations, Nordvik Region (landscape/models/nordvik_region.yaml) and Hansa Mutual (landscape/models/hansa_mutual.yaml), whose canonical profiles are in docs/EXAMPLE_ORGANISATIONS.md. They are not a measurement of any real organisation and must never be presented as data, compared with data in the same chart or used to support a factual claim (docs/STYLE_GUIDE.md, data and caption rule 3). Units of the two illustrative landscapes per autonomy level and per usage mode, state at 2027-01-01. Compiled 2026-11-07 from repository content; no external source. Licence: repository content; the models and the counts are the author's own illustrative material, with no external source. Reproduction in this book: the chart is captioned "Illustrative" and never carries a Source: line naming a publisher.

Caveats recorded at retrieval:

Illustrative, not data. Every figure drawn from this dataset is captioned "Illustrative" and the spec's landscape block carries "illustrative": true
The date matters. The counts change over 2027 as the timeline entries take effect; the file holds the start-of-year state only. Run the script with --date 2027-12-31 for the end-of-year state
Paused and pilot units count. A paused unit (Nordvik Region's NR-TL) is still a unit in the landscape and is counted at its level; only units not yet present are excluded

`microsoft_work_trend_index_2025`

Microsoft Corporation (WorkLab): vendor survey and vendor telemetry. Survey fieldwork by Edelman Data x Intelligence. Microsoft Work Trend Index Annual Report 2025 ("2025: The year the Frontier Firm is born"), with the 2026 edition. Marked as a vendor survey in every caption. Retrieved 2026-11-07 from https://www.microsoft.com/en-us/worklab/work-trend-index/2025-the-year-the-frontier-firm-is-born. Licence: No Creative Commons licence; © Microsoft, under Microsoft's website terms of use.

Caveats recorded at retrieval:

Vendor marketing research: online panel of knowledge workers (self-reported), "leaders" defined by job title; "Frontier Firm" is a Microsoft construct. Agent-adoption percentages describe expectations and self-reported use among respondents, not measured deployment
The 2025 telemetry excludes EU tenants; the 2026 Copilot analysis covers North America only: relevant for a book about European organisations
The 2026 survey only includes workers who already use AI (20,000), so its percentages are shares of AI users, not of the workforce

`oecd_aim_incidents`

OECD (OECD.AI Policy Observatory), AIM developed by the OECD.AI expert group on AI incidents with the support of the Patrick J. McGovern Foundation; data provided by Event Registry, "Data processing powered by Microsoft Azure". "AIM: AI Incidents and Hazards Monitor: Automated media discourse monitor of AI incidents and hazards (Beta)". Retrieved 2026-11-07 from https://oecd.ai/en/incidents. Licence: The methodology page states: "Your use of the OECD AI incidents and hazards monitor ... ('AIM') is subject to the terms and conditions found at www.oecd.org/termsandconditions" (https://www.oecd.org/termsandconditions/). OECD terms and conditions permit reproduction of OECD data and text for non-commercial and, with acknowledgement, commercial purposes subject to the conditions stated there; the counts here are AIM's own aggregates (not third-party article text), so reproducing derived figures in a published book with the attribution "OECD.AI AI Incidents and Hazards Monitor (AIM), retrieved 7 November 2026" is consistent with those terms, but the author should check the OECD terms page before print because AIM is labelled Beta and the OECD states that "The information displayed in the AIM should not be reported as representing the official views of the OECD or of its member countries." The monitor is not published under a Creative Commons licence. No public API documentation exists; the endpoint is the one the website itself uses and may change.

Caveats recorded at retrieval:

Method: AIM detects events automatically from "reputable international news outlets" via Event Registry (over 150,000 news articles daily), classifies them with AI ("[AI generated]" labels on every record), and since November 2024 applies the OECD definitions of AI incident (actual harm) and AI hazard (plausible harm). Historical events before November 2024 "are currently being retrospectively annotated using the same methodology". Counts are therefore counts of news-reported events, not of harms, and depend on media coverage, language coverage (English, French, German, Arabic and others), and the classifier
The OECD's own landing page notes that incidents "have actually gone down as a share of all AI news", so absolute growth partly tracks the growth of AI news; use ai_event_counts as the denominator
Country attribution is the location tagged for the event; many events have no country
The dataset is updated continuously; totals will differ on any later retrieval. All numbers here are as of the 2026-11-07 retrieval

`pew_ai_public_experts_2025`

Pew Research Center, Washington, D.C. (nonpartisan, nonprofit; not vendor-commissioned). "How the U.S. Public and AI Experts View Artificial Intelligence." Pew Research Center, Washington, D.C. (April 3, 2025) https://www.pewresearch.org/internet/2025/04/03/how-the-us-public-and-ai-experts-view-artificial-intelligence/ . Authors: Colleen McClain et al. Retrieved 2026-11-07 from https://www.pewresearch.org/internet/2025/04/03/how-the-us-public-and-ai-experts-view-artificial-intelligence/. Licence: "In furtherance of the Center's mission to inform the public debate on important issues, you may access, print, copy, reproduce, cite, link, display, download, distribute, broadcast, transmit, publish, license, transfer, sell, modify, create derivatives of, or otherwise exploit the Content, provided that all copies display all copyright and other applicable notices to the extent such notices are contained in such Content and provided further that you do not use the Content in any manner that implies, suggests, or could otherwise be perceived as attributing a particular policy or lobbying objective or opinion to the Center, or as a Center endorsement of a cause, candidate, issue, party, product, business, organization, religion or viewpoint. You must also provide proper attribution to the Center in connection with your use of any Content with express reference to the Center in accordance with the citation below." Citation format (Section 2): "'Report Title.' Pew Research Center, Washington, D.C. (Publication date) URL." Note: the terms allow reproduction including in sold works, provided attribution and no implied endorsement; they are broader than "non-commercial only". For survey datasets (not used here) the additional disclaimer "Pew Research Center bears no responsibility for the analyses or interpretations of the data presented here..." applies.

Caveats recorded at retrieval:

U.S. public only; no European respondents. The expert sample is a convenience sample (conference authors/presenters), unweighted, and Pew warns it is not representative of all AI experts
Split-form items (AIFUTRIMPCT) have half the sample size

`pew_global_views_ai_2025`

Pew Research Center, Washington, D.C. (nonpartisan, nonprofit; not vendor-commissioned). "How People Around the World View AI." Pew Research Center, Washington, D.C. (October 15, 2025) https://www.pewresearch.org/global/2025/10/15/how-people-around-the-world-view-ai/. Retrieved 2026-11-07 from https://www.pewresearch.org/global/2025/10/15/how-people-around-the-world-view-ai/. Licence: as quoted in ../pew_ai_public_experts_2025/SOURCE.md (https://www.pewresearch.org/about/terms-and-conditions/): content may be reproduced, published and sold with attribution in Pew's citation format and without implying Pew endorsement. Figures may be reproduced in a published book with that attribution.

Caveats recorded at retrieval:

Modes differ between countries (telephone in Europe, face-to-face elsewhere); U.S. items come from two different ATP waves
No question on AI use at work in this release; the topline says omitted questions "have either been previously released or will be released in future reports"

`series_diagram_survey`

The author's own repositories (Gunnar Nordqvist, Geonitab). This is not an external statistic: it is a count of the Mermaid diagram files in the seven volumes of The Transformed Architecture, made for docs/FOUNDATIONS.md ("Survey of the diagrams of the seven volumes"). Survey of the 284 Mermaid diagrams of the seven volumes: Mermaid type, drawing direction and recurring subjects. Compiled 2026-11-07 from repository content; no external source. Licence: the author's own repositories. The seven repositories and their diagrams are the author's work; counting them needs no third-party licence. Reproduction in this book: reproducing these counts in the book is the author reproducing a measurement of his own work; the caption says "Source: the author's survey of the diagrams of the seven volumes (2026)".

Caveats recorded at retrieval:

Keyword matching is coarse. It finds diagrams whose labels name the subject, not every diagram that implies it; a label such as "approval" counts as decision rights wherever it appears. The subject counts are exact for what the files say and approximate for what the diagrams mean
Direction is read from the first statement. TB and TD are two spellings of the same direction and are counted together
The survey is a snapshot of the working copies on 2026-11-07. Diagrams added to the volumes after that date are not counted
Not a statistic about the world. The dataset describes seven books by one author; it supports no claim about practitioners in general

`stackoverflow_dev_survey_2024`

Stack Overflow (Stack Exchange Inc.); self-selected online survey of Stack Overflow users (company/community survey, not a probability sample; not commissioned by a third party). 2024 Stack Overflow Developer Survey. Open 2024-05-20 to 2024-06-20, released 2024-07-24 (archive README). Stack Overflow: "In May 2024, over 65,000 developers responded to our annual survey about coding, the technologies and tools they use and want to learn, AI, and developer experience at work." (https://survey.stackoverflow.co/2024/). Stack Overflow's own published AI headline: "76% of all respondents are using or are planning to use AI tools in their development process this year, an increase from last year (70%). Many more developers are currently using AI tools this year, too (62% vs. 44%).". Retrieved 2026-11-07 from https://survey.stackoverflow.co/2024/. Licence: identical to 2025: ODbL 1.0 for the database, DbCL 1.0 for contents; "Attribution: Stack Overflow Developer Survey, Stack Exchange Inc." Aggregates may be reproduced in a book with that attribution.

Caveats recorded at retrieval:

As for 2025: convenience sample; Stack Overflow's website figures use their own bases (e.g. their "62% currently using" is close to but not identical with the 61.84% of item respondents here). Task wording changed between 2024 and 2025 ("Debugging and getting help" → "Debugging or fixing code"; "Learning new concepts..." and "Creating or maintaining documentation" added), so the AITool series is not strictly comparable across years

`stackoverflow_dev_survey_2025`

Stack Overflow (Stack Exchange Inc.). Self-selected online survey promoted to Stack Overflow users; not commissioned by a third party but a company publication (treat as a vendor/community survey, not a probability sample). 2025 Stack Overflow Developer Survey. Survey open 2025-05-29 to 2025-06-23, results released 2025-07-29 (archive README). Stack Overflow's own description: "In its fifteenth year, Stack Overflow received over 49,000+ responses from 177 countries across 62 questions focused on 314 different technologies, including new focus on AI agent tools, LLMs and community platforms." (https://survey.stackoverflow.co/2025/). Marked as a vendor survey in every caption. Retrieved 2026-11-07 from https://survey.stackoverflow.co/2025/. Licence: "Published by Stack Overflow under the Open Database License (ODbL) 1.0 (https://opendatacommons.org/licenses/odbl/1-0/); individual cell contents under the Database Contents License (DbCL) 1.0 (https://opendatacommons.org/licenses/dbcl/1-0/). Attribution: Stack Overflow Developer Survey, Stack Exchange Inc." The repository-level LICENSE.md is Apache 2.0 (for the repository code). Aggregated figures may be reproduced in a published book with that attribution; a redistributed derived database must stay under ODbL.

Caveats recorded at retrieval:

Non-probability, self-selected sample of Stack Overflow visitors; heavily skewed to professional developers in the U.S., Germany, India and the UK. "EU27" here is a geographic filter on this convenience sample, not a representative European figure
Partial responses are included in the file; later questions have smaller bases (e.g. AISelect n = 33,720 of 49,191)

`stanford_ai_index_2025`

AI Index Steering Committee, Institute for Human-Centered Artificial Intelligence (HAI), Stanford University. Artificial Intelligence Index Report 2025 (PDF metadata title), April 2025, 457 PDF pages. Suggested citation (PDF p.8): Nestor Maslej et al., "The AI Index 2025 Annual Report," AI Index Steering Committee, Institute for Human-Centered AI, Stanford University, Stanford, CA, April 2025. https://doi.org/10.48550/arXiv.2504.07139. Retrieved 2026-11-07 from https://doi.org/10.48550/arXiv.2504.07139. Licence: PDF p.8: "The AI Index 2025 Annual Report by Stanford University is licensed under Attribution-NoDerivatives 4.0 International." (CC BY-ND 4.0, https://creativecommons.org/licenses/by-nd/4.0/). The same page invites readers to use the public data: "Public Data and Tools: The AI Index 2025 Report is supplemented by raw data and an interactive tool. We invite each reader to use the data and the tool in a way most relevant to their work and interests. Raw data and charts: The public data and high-resolution images of all the charts in the report are available on Google Drive.".

Caveats recorded at retrieval:

AIID incident counts depend on reporting and editorial throughput and are revised retroactively (see aiid_incidents/, where the 2026-11-02 snapshot gives 313 incidents dated 2024 against the Index's 233)
Corporate adoption figures are from McKinsey's "State of AI" online survey of its own respondent panel (vendor survey, self-reported, over-representing large firms and AI-engaged respondents); the Index relays them
Public-opinion figures are Ipsos online panel surveys (relayed); "Great Britain" is Ipsos's label for the UK sample
Investment data (Quid) count disclosed private deals and are revised between editions
A 2026 edition exists (see stanford_ai_index_2026/); its 2025 incident count is 362

`stanford_ai_index_2026`

AI Index Steering Committee, Stanford HAI. Artificial Intelligence Index Report 2026, April 2026, 425 PDF pages. Suggested citation (PDF p.6): "The AI Index 2026 Annual Report," AI Index Steering Committee, Institute for Human-Centered AI, Stanford University, Stanford, CA, April 2026. https://doi.org/10.48550/arXiv.2606.15708. Retrieved 2026-11-07 from https://doi.org/10.48550/arXiv.2606.15708. Licence: PDF p.6: "The AI Index 2026 Annual Report by Stanford University is licensed under Attribution-NoDerivatives 4.0 International." Same implications as for the 2025 edition.

`test_answers`

This repository. The table is compiled from the other datasets in data/; it contains no value that is not in one of their tidy.csv files or derived from two of them by a stated arithmetic step. It exists so that Chapter 1's test (chart_test_answers) can be drawn from one file and so that every answer is traceable to its dataset. The fourteen test questions of Chapter 1 with their three options and the measured answer. Compiled 2026-11-07 from repository content; no external source.

Caveats recorded at retrieval:

Compiled, not retrieved. Regenerate with the script whenever an underlying dataset changes; do not edit by hand
Question 4 is a derived ratio of two published shares; Question 13's use figure is a sum of three published option shares. Both are labelled as derived in method
Question 9 is a projection of the European Commission, not an observation
Question 13 rests on a self-selected online sample (Stack Overflow); Question 14 on a probability telephone survey (JRC AIM-WORK); the chapter says so when it gives the answers

Sources

Nordqvist, G. (2025). Architecture as Code. The Transformed Architecture, Book 1, aac.geon.se. (1, Nordqvist, 2025)
Nordqvist, G. (2026). AI Governance: Governing Artificial Intelligence in European Enterprises and Public Authorities. The Transformed Architecture, Book 2, aig.geon.se. (2, Nordqvist, 2026)
Nordqvist, G. (2026). AI Platform and Security: Building and Securing AI Platforms in European Enterprises and Public Authorities. The Transformed Architecture, Book 3, aips.geon.se. (3, Nordqvist, 2026)
Nordqvist, G. (2026). AI Red Team: AI as an Active Part of Cyber Defence in European Enterprises and Public Authorities. The Transformed Architecture, Book 4, aired.geon.se. (4, Nordqvist, 2026)
Nordqvist, G. (2026). AI Threat: Are the Dystopias Coming True? Evidence, Risk and Response for European Enterprises and Public Authorities. The Transformed Architecture, Book 5, aithreat.geon.se. (5, Nordqvist, 2026)
Nordqvist, G. (2026). The Transformed Security Architect: Role, Craft and Career in the Age of AI in European Enterprises and Public Authorities. The Transformed Architecture, Book 6, tsa.geon.se. (6, Nordqvist, 2026)
Nordqvist, G. (2026). AI Boss: When Artificial Intelligence Manages People in European Enterprises and Public Authorities. The Transformed Architecture, Book 7, aiboss.geon.se. (7, Nordqvist, 2026)
Rosling, H., Rosling, O. and Rosling Rönnlund, A. (2018). Factfulness: Ten Reasons We're Wrong About the World and Why Things Are Better Than You Think. London: Sceptre. (8, Rosling et al., 2018)
Eurostat (2026). Artificial intelligence by size class of enterprise. Online data code isoc_eb_ai. (9, Eurostat, 2026)
Eurostat (2026). Cloud computing services by size class of enterprise. Online data code isoc_cicce_use. (10, Eurostat, 2026)
Eurostat (2026). Security incidents and consequences by size class of enterprise. Online data code isoc_cisce_ic. (11, Eurostat, 2026)
Eurostat (2026). Employed information and communications technology (ICT) specialists. Online data code isoc_sks_itspt. (12, Eurostat, 2026)
Eurostat (2026). Enterprises that recruited or tried to recruit ICT specialists by size class of enterprise. Online data code isoc_ske_itrcrs. (13, Eurostat, 2026)
Eurostat (2026). Individuals - use of generative AI tools. Online data code isoc_ai_iaiu. (14, Eurostat, 2026)
European Commission, Directorate-General for Communications Networks, Content and Technology (2026). Digital Decade DESI dashboard: Digital Decade EU-level trajectories and Digital Economy and Society Indicators. (15, European Commission, 2026)
Gapminder Foundation (2017). Gapminder Misconception Study 2017. (16, Gapminder Foundation, 2017)
Maslej, N. et al. (2025). The AI Index 2025 Annual Report. AI Index Steering Committee, Institute for Human-Centered AI, Stanford University. (17, Maslej et al., 2025)
Sajadieh, S. et al. (2026). The AI Index 2026 Annual Report. AI Index Steering Committee, Institute for Human-Centered AI, Stanford University. (18, Sajadieh et al., 2026)
DeBellis, D. et al. (2024). Accelerate State of DevOps Report 2024. DORA, Google Cloud. (19, DeBellis et al., 2024)
DeBellis, D. et al. (2025). State of AI-assisted Software Development. DORA, Google Cloud. (20, DeBellis et al., 2025)
Silverthorne, V. and Hendrick, S. (2025). Cloud Native 2024: Approaching a Decade of Code, Cloud, and Change. CNCF Annual Cloud Native Survey 2024, The Linux Foundation. (21, Silverthorne & Hendrick, 2025)
GitHub (2024). Octoverse: AI leads Python to top language as the number of global developers surges. GitHub Blog. (22, GitHub, 2024)
GitHub (2025). Octoverse: A new developer joins GitHub every second as AI leads TypeScript to #1. GitHub Blog. (23, GitHub, 2025)
GitHub (2026). Octoverse: Commits, pushes, reach new highs in 2026 as developers build with agents. GitHub Blog. (24, GitHub, 2026)
Microsoft (2025). 2025: The year the Frontier Firm is born. Work Trend Index Annual Report, Microsoft WorkLab. (25, Microsoft, 2025)
Microsoft (2026). 2026 Work Trend Index Annual Report: Agents, human agency, and the opportunity for every organization. Microsoft WorkLab. (26, Microsoft, 2026)
Stack Overflow (2024). 2024 Stack Overflow Developer Survey. Stack Exchange Inc. (27, Stack Overflow, 2024)
Stack Overflow (2025). 2025 Stack Overflow Developer Survey. Stack Exchange Inc. (28, Stack Overflow, 2025)
European Commission, Directorate-General for Employment, Social Affairs and Inclusion (2025). Artificial Intelligence and the future of work. Special Eurobarometer 554. (29, European Commission, 2025)
European Commission, Directorate-General for Communications Networks, Content and Technology (2025). The Digital Decade 2025. Special Eurobarometer 566. (30, European Commission, 2025)
Ipsos (2025). The Ipsos AI Monitor 2025: A 30-country Ipsos Global Advisor Survey. Ipsos. (31, Ipsos, 2025)
McClain, C., Kennedy, B., Gottfried, J., Anderson, M. and Pasquini, G. (2025). How the U.S. Public and AI Experts View Artificial Intelligence. Pew Research Center. (32, McClain et al., 2025)
Poushter, J., Fagan, M. and Corichi, M. (2025). How People Around the World View AI. Pew Research Center. (33, Poushter et al., 2025)
Gillespie, N., Lockey, S., Ward, T., Macdade, A. and Hassed, G. (2025). Trust, attitudes and use of artificial intelligence: A global study 2025. The University of Melbourne and KPMG. (34, Gillespie et al., 2025)
Responsible AI Collaborative (2026). AI Incident Database. incidentdatabase.ai. (35, Responsible AI Collaborative, 2026)
Organisation for Economic Co-operation and Development (2026). AIM: AI Incidents and Hazards Monitor. OECD.AI Policy Observatory. (36, OECD, 2026)
European Union Agency for Cybersecurity (2024). ENISA Threat Landscape 2024. ENISA. (37, ENISA, 2024)
European Union Agency for Cybersecurity (2025). ENISA Threat Landscape 2025. ENISA. (38, ENISA, 2025)
European Parliament and Council of the European Union (2024). Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union. (39, European Union, 2024)
European Parliament and Council of the European Union (2026). Regulation (EU) 2026/1744 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 as regards the simplification of the implementation of harmonised rules on artificial intelligence (Digital Omnibus on AI). Official Journal of the European Union. (40, European Union, 2026)
European Parliament and Council of the European Union (2022). Decision (EU) 2022/2481 establishing the Digital Decade Policy Programme 2030. Official Journal of the European Union. (43, European Union, 2022)
European Commission (2011). Commission Decision of 12 December 2011 on the reuse of Commission documents (2011/833/EU). Official Journal of the European Union. (45, European Commission, 2011)
Eurostat (n.d.). Copyright notice and free re-use of data. Eurostat help pages. (48, Eurostat, n.d.)
European Commission (n.d.). Legal notice. commission.europa.eu. (49, European Commission, n.d.)
McGregor, S. (2021). Preventing Repeated Real World AI Failures by Cataloging Incidents: The AI Incident Database. Proceedings of the AAAI Conference on Artificial Intelligence. (50, McGregor, 2021)
Gonzalez Vazquez, I., Fernandez Macias, E., Wright, S. and Villani, D. (2025). Digital Monitoring, Algorithmic Management and the Platformisation of Work in Europe. European Commission, Joint Research Centre, JRC143072. (51, Gonzalez Vazquez et al., 2025)
European Commission (2026). Proposal for a Regulation establishing a framework of measures for strengthening Europe's cloud and AI ecosystem (Cloud and AI Development Act). COM(2026) 502. (46, European Commission, 2026)
European Commission (2026). Commission Staff Working Document: Impact Assessment Report accompanying the proposal for the Cloud and AI Development Act. SWD(2026) 502. (47, European Commission, 2026)
Synergy Research Group (2022). European Cloud Providers Continue to Grow but Still Lose Market Share. Press release, 27 September 2022. (52, Synergy Research Group, 2022)
Synergy Research Group (2025). European Cloud Providers' Local Market Share Now Holds Steady at 15%. Press release, 24 July 2025. (53, Synergy Research Group, 2025)