Lagging Lagging Behind the Speed — Statecraft

Summary

In March 2026, 1,392 algorithms were registered in the Dutch national Algorithm Register, published by just over five hundred government organisations. The Dutch government is estimated to use sixteen thousand of them. Only two per cent of the algorithms that directly affect data subjects has been published. A three-year-old parliamentary motion to make registration legally mandatory is still awaiting implementation. The Human Rights and Algorithms Impact Assessment (IAMA) was revised in February 2026 to Version 2 and aligned with Article 27 of the European AI Act, but is still not a stand-alone law. The Implementation Act for the AI Regulation entered public consultation on 20 April 2026. The European AI Act, which fully enters into force for high-risk systems on 2 August 2026, was itself supplemented by the European Commission in November 2025 with a postponement proposal because the accompanying harmonised standards will not be ready in time.

This signature essay names the pattern that runs through all five preceding doorwerkingen (knock-on effects) and that adds to the demographic blindness of Paper 6 its complementary counterpart. Where the previous paper documented that the apparatus fails in the face of futures it has long known, this paper documents that the apparatus also fails in the face of presents it cannot keep up with. The diagnosis is administrative, not technical. The problem is not a lack of knowledge. It is a speed differential between exogenous time, in which technology unfolds in cycles of months, and institutional time, in which governance, jurisprudence and accountability unfold in cycles of years to decades.

The handling perspective does not require slowing down exogenous speed, because no single state can do that alone. It requires accelerating institutional adaptive capacity, modelled on states that have begun treating the speed differential as a design problem.

§ 01 An Algorithm Register that keeps falling behind

The Dutch national Algorithm Register was launched in December 2022. The trigger was the childcare benefits affair and the realisation that a significant part of the injustice done to citizens had been delivered along algorithmic lines, without any citizen or supervisor being able to see in advance which models were being used in which decision chains. The idea was that openness about algorithms would form the first link in a chain of public oversight that, via supervision, jurisprudence and legislation, would ultimately result in democratically controllable automation. A reasonable thought, delivered at a reasonable moment.

In the years that followed, the register moved forward at the initial pace of a ferry. By the end of 2023, there were 264 entries.¹ In early 2025, more than two years after the start, there were 730. In March 2026, there were 1,392. At first sight, a fivefold increase in two years. On closer inspection, a fraction of what it should be. Open State Foundation calculated earlier that only about two per cent of the algorithms with direct impact on data subjects had been included in the register at any point.² The Dutch government is estimated to use sixteen thousand algorithms in active operation. The parliamentary motion to make registration mandatory has been on the table since 2022. In the parliamentary debate of January 2025, the then State Secretary for Digitalisation declared he would discuss it in the Council of Ministers. By April 2026, that conversation has not resulted in implementation.

The pace of the apparatus, in short, is not a problem of capacity. It is a problem of ownership. No minister, on taking office, weighs the non-publication of algorithms in the damage risk of their portfolio. No secretary-general steers on the percentage of high-impact algorithms in the register as a KPI for their directorate. No internal audit of an executive agency measures whether its own algorithms are due for publication. The register exists. It is an instrument without a hand to operate it.

At the same time, three shifts are unfolding at the exogenous speed of technological development that will render the register obsolete within two years in its current form. The first is the rise of large language models in Dutch executive work, in the form of caseworker assistants, transcription services in youth care, summary engines in permitting, and chatbots in the front line of citizen contact. They are rarely registered because they are not perceived as “algorithms” by the procurement line. The second is the rise of predictive analysis in selection rooms, fraud detection and enforcement, where models are often kept below the threshold of an impactful algorithm description by the using organisation itself. The third is the external supplier architecture, where the Dutch government increasingly does not build but procures algorithms, thereby pushing technical, legal and ethical responsibility into a supplier relationship that neither the register nor the Senior Civil Service can adequately disentangle.

What was a reasonable first-aid intervention in 2022 is in 2026 an instrument overtaken by reality itself. Not because the instrument was poorly designed, but because the speed differential between instrument design and exogenous development is structurally larger than a register can close. The Dutch Data Protection Authority noted in March 2026, in its sixth Report on AI & Algorithms in the Netherlands, that four out of nine indicators on the AI Impact Barometer are now red, where previously two were.³ Chair Wolfsen formulated it in the same report in seven words: five years after the benefits scandal, the lessons are clear but the follow-up is lagging.

§ 02 The algorithmic dimension of the childcare benefits affair

In October 2020, the newspaper Trouw published the story of the Tax Authority’s risk classifications, which automatically sorted childcare benefit applications for fraud suspicion partly on the basis of nationality and family name.⁴ In the years that followed, parts of this automatic sorting were declared by the administrative court to be in breach of the law, partly on the basis of the prohibition of discrimination in Article 1 of the Constitution and Article 14 of the European Convention on Human Rights. The parliamentary committee on the childcare benefits affair concluded in December 2020, in Ongekend onrecht (Unprecedented Injustice), that administrative responsibility had failed at every level.⁵ The parliamentary inquiry committee on Fraud Policy and Service Delivery presented on 26 February 2024 its final report Blind voor mens en recht (Blind to Human Beings and the Law), which explicitly identified the algorithmic dimension as constitutive of the injustice. The cabinet responded on 5 December 2024.

What rarely became the subject of diagnosis in this sequence of reports is that the algorithmic sorting, before being made public in 2020, had grown over the years alongside the fraud-hunting doctrine without any institutional link keeping pace with that growth. The system was put into production in 2014. It was further developed in 2015 and 2016. It developed fraud scores based on self-learning data and received an extension in 2018, in which the scores were recalibrated on the basis of earlier outcomes. The Data Protection Authority, the Government Audit Service, the National Ombudsman, the Court of Audit: none of them was ready, at the time of further development, with a supervisory framework capable of keeping up with the speed of model adjustment. The GDPR framework only became operationally effective when the system had already taken root. The court only came into the picture when citizens stood at the gate by the hundreds.

This is not the incompetence of a specific official. It is temporal asymmetry. The technical cycle in which a scoring engine is recalibrated is a matter of weeks. The cycle in which a supervisor can prepare a methodological assessment is a matter of months to years. The cycle in which a court can develop a systematic test against rights principles is a matter of years to a decade. The cycle in which a law is adapted to a new technological reality is a matter of a parliamentary term or longer. Four different clocks, measured in four different seconds, racing against a fifth clock that ticks in milliseconds.

In Mark Moore’s Strategic Triangle, this is a classic disturbance of operational capacity, which is missing here not because of capacity shortage but because of a speed differential. Public value, a just approach to fraud, was unambiguous. Political legitimacy of fraud control was even heightened throughout the period 2014-2019. What was lacking was operational capacity at the speed of change: the ability to develop, test and correct, within a reasonable time, a workable, legally tenable and ethically defensible model. Not the direction of the triangle, but its rotational speed.

§ 03 The Tax Authority data vault as technologically facilitated dissociation

The Statecraft paper De architectuur van het zwijgen (The Architecture of Silence, April 2026) described how the Tax Authority’s data vault could store sixty-four million documents for seven years without any supervisor, state secretary or parliamentary committee seeing its content. The diagnosis was institutional: the chain of Government Audit Service, Data Protection Authority, Ombudsman, two successive state secretaries and a parliamentary inquiry committee collectively allowed a file to evaporate even though every link operated formally correctly.⁶

What that paper did not develop is that the specific scale of the evaporation was made possible technologically. In the paper-based administrative practice of twenty years earlier, sixty-four million documents would have been a physical problem. They could not have been stored without a building, and no building without a custodian. The storage capacity of a digital vault is boundless and silent. What disappears into it takes up no space and makes no sound. The custodian is not a person with a key ring but a server park administrator whose work succeeds as long as nothing happens. That is a fundamentally different condition. The apparatus had always had a tendency not to see things it found uncomfortable. Digital technology has scaled that tendency into an industrial possibility.

That explains the difference in scale between classical organisational dissociation and its digitally facilitated version. In a paper archive, an unmanaged file would draw attention from time to time through physical presence. In a digital storage system without lifecycle management, a file can remain untouched for seven years without a bell ringing anywhere. The institutional capacity to counter “silent disappearance” was historically built on physical counter-incentives that have not been replaced in the digital design by anything equivalent. Audit cycles, data lifecycle management and statutory destruction periods presuppose active stewardship. When stewardship is left to an infrastructure that has no agency of its own, the apparatus succeeds in institutionalising forgetting.

The problem is therefore not that technology has changed organisations. The problem is that the institutional processing of technology lags behind the properties of that technology itself. In a paper archive, discipline was imposed by the material itself. In a digital archive, discipline must be designed explicitly, and that designing has not happened at the speed at which digital storage has been rolled out. The architecture of silence operates more loudly in a digital substrate than in a paper one, precisely because it has become quieter.

§ 04 The attention economy of the executive civil servant

In Limbic Literacy (House of Viridian, draft 2026), the attention economy is described as an industrial sector in which specialised designers, data scientists and behavioural experts work daily to calibrate stimuli that play on the user’s limbic system at a speed that consciousness cannot keep up with.⁷ The prefrontal cortex responds in two hundred to five hundred milliseconds; the limbic system in twenty to fifty. The different reaction speeds are not gradual but categorical. What that book works out for private life applies, in a modified but related way, to the work environment of the contemporary executive civil servant.

A Tuesday morning case

In an interim assignment in the social domain of a municipality with just over a hundred thousand inhabitants, in the first month I spent a morning shadowing a Wmo (Social Support Act) caseworker. The administrative reason was that the department had a persistent absenteeism problem, and the director wanted to understand why a team of thirty caseworkers, with formative scope for thirty-six, structurally reported under-capacity. Productivity figures per caseworker had been stable since 2018. Caseload was within VNG (Association of Dutch Municipalities) guidelines. Employee satisfaction was middling, not alarming. On paper, nothing was wrong. In practice, a third of the establishment was chronically in soft drop-out: short absences, self-taken extra days, slow turnaround of kitchen-table conversations.

The morning was instructive. Between eight and quarter past nine, the caseworker had opened six different systems: a municipal case file system for the client dossier, an adjacent system for financial decisions, the CAK portal for personal-contribution calculations, a regional data hub for care providers, an inbox that produced eleven new messages in thirty minutes (four internal, three from care providers, two from clients, one from a GP, one from Veilig Thuis), and a Teams environment in which the department chat added a new line every two to three minutes. On the mobile phone, additional notifications came in from a crisis-case alert system, a care coordination app for the neighbourhood team, and a personal calendar. On the desk lay a paper tab with the day’s order, updated once between the first and second phone calls. By eleven o’clock, after a home visit that had lasted forty minutes and been interrupted in the car by three incoming calls, I asked the caseworker where she stood with the morning work she had prioritised at eight. The answer was: nowhere.

This is not an incident. In 2026 it is the main mode of executive work. On the workstation of a Wmo caseworker, a permit officer, a youth-care officer or a customer-contact-centre employee, an average of six to ten parallel systems are running. The caseworker at a municipality works at a desk with thirty open tabs, an inbox that produces a notification every five minutes, a chat function that orchestrates collective work traffic, a case portal that brings five systems into view per client, and increasingly an AI assistant that pushes out pre-selected recommendations. The permit officer works through the same infrastructure, with on top of that a schedule of consultation duties spanning the whole policy spectrum. The youth-care worker receives notifications on his phone from four different alert systems. The integral attention economy of executive work has been redesigned in fifteen years along lines drawn not by public administration scholars, but by product designers of software vendors whose business model depends on what they call attention engagement.

Temporal asymmetry in its microscopic form

In that redesign sits the temporal asymmetry of signature II in its most microscopic form. The product designer iterates in two-week sprints. The procurement procedure is a matter of months. Changing standard work processes in an executive organisation is a matter of one to two years. Changing the Police Data Act, the General Administrative Law Act or the Open Government Act is a matter of a parliamentary term. The speed differential means that no executive civil servant can look back on a working environment designed at their own production-relevant pace. The environment is delivered at the supplier’s pace. The work is evaluated at the legislator’s pace. The employee sits in between.

What the Dutch data show

The effect of this is measurable on three levels.

At the macro level, the National Working Conditions Survey by TNO and Statistics Netherlands has documented since 2015 a steadily rising perceived workload and a doubling of burnout complaints. The percentage of employees with burnout complaints rose from thirteen per cent in 2015 to twenty per cent in 2024, with an interim peak of twenty per cent in 2022 and a slight dip in 2023.⁸ Average screen time is four point four hours per day; forty-four per cent works long hours (six hours or more) at a screen. For the public administration sector, the absenteeism rate in 2024 stood at six point one per cent, structurally above the national average of five point two per cent, and rose further in the first quarter of 2025 to six point nine per cent. ArboNed reports that public administration, healthcare and education form the top three sectors for stress-related absenteeism in 2024.⁹ According to the Arbobalans 2024, continued-pay costs of work-related sickness rose from five point one billion euros in 2015 to eight point three billion in 2023, of which four point nine billion is directly attributable to psychosocial workload.

At the municipal level, the Personnel Monitor Municipalities 2024 confirms the pattern. Sickness absence at municipalities reached six point seven per cent in 2024 (G4 cities at eight per cent). Nine out of ten municipalities call workload a current HR issue, and seventy-two per cent of municipalities took active workload measures in 2024, up from sixty-seven per cent in 2023.¹⁰ Occupational physicians cite workload and stress as a cause in sixty-three per cent of long-term absence cases. External hire as a percentage of payroll came in at seventeen point five per cent: in an executive domain where intra-municipal continuity is by definition the baseline, a strikingly high figure.

At the level of the working day itself, the measurement has not yet been made in the Netherlands. Outside the Netherlands, it has. Gloria Mark documents in Attention Span (2023), based on five studies of knowledge workers, that the average uninterrupted focus duration on one screen has decreased from one hundred fifty seconds in 2004 to seventy-five in 2012 to an average of forty-seven, median forty, in her most recent measurements. The refocus time after an interruption, where the original work is taken up again, averages twenty-five minutes.¹¹ Microsoft documented in its Work Trend Index 2025 that Microsoft 365 users worldwide are interrupted on average every two minutes by a meeting, email or notification, and process on average one hundred fifty-three Teams messages and one hundred seventeen emails per day.¹²

The empirical blank spot

At this point, an honest methodological remark is in order. A Dutch measurement of attention fragmentation among civil-service knowledge workers, at the level of Mark’s screen tracking, does not exist. The NEA does not break workload down to the level of digital interruptions. The Personnel Monitor does not break workload down by functional area (social domain, permitting, customer contact, youth care). The 2024 ICTU Work Survey does show that only fifty-two per cent of civil servants feel engaged with their organisation and that fifty-one per cent believe underperformance is tolerated, but this measurement too is not disaggregated by functional domain. The academic literature on screen-level bureaucracy is internationally stronger than nationally: Bovens and Zouridis showed in 2002 how street-level bureaucrats are increasingly being replaced by system-level bureaucrats, and Soares, Grimmelikhuijsen and Meijer published an ethnographic study in 2024 of an online-fraud team at the Dutch police in which the cognitive switch cost between screens was empirically documented.¹³ What is missing is a Dutch macro measurement that closes the claim “digital attention load in executive work has steadily increased, with measurable consequences” in quantitative form. For version 2 of this paper, the triangle of evidence is therefore: international micro (Mark, Leroy, Tarafdar), Dutch academic meso (Bovens, Zouridis, Soares, Grimmelikhuijsen, Meijer), and Dutch statistical macro (NEA, Personnel Monitor, CBS, ArboNed). What lies between meso and macro, a direct measurement among civil-service knowledge workers in the Netherlands, is a blank spot which this diagnosis itself helps to put on the agenda.

Change colours and Aiki action

In De Caluwé and Vermaak’s change colours, this is an issue that does not sit in the blue planning-based column where it is normally addressed. It is in significant part red, namely a question of work experience and human scale, and in another part white, namely a question of self-organisation by professional groups designing their own attention environment. The blue planning approach to workload via roster programming and establishment expansion provides no solution if the cause lies in daily attention calibration. Aiki action on this issue would mean not frontally resisting the energy of the supplier architecture, but redirecting it towards a design register in which executive work is built up according to human attention rhythms. Not a new law on screen time, but a new norm for procurement specifications in which attention discipline of the designed instrument is a measurable acceptance criterion.

In the team I shadowed that Tuesday morning, the first month of intervention did not yield a redesign of the software stack, because that lies in a procurement that a local interim manager cannot reopen. What it did yield was a team agreement to switch off Teams notifications between nine and eleven, an inbox opened only at fixed moments, and a case-management approach that no longer opened six systems per client in succession but one summary overview screen, built by a seconded information analyst within three weeks from the existing system APIs. These are micro-interventions with measurable outcomes (productivity per caseworker rose by about fifteen per cent in two months; absenteeism began to tilt slightly), but they confirm the paper’s claim: the problem does not sit in individual work discipline, but in a working environment that no one designs at system level for attention discipline.

§ 05 Three domains where exogenous speed escapes institutional grip

Three domains show the speed differential in its most policy-relevant form.

Privacy. The GDPR is a European framework from 2016 that became effective in 2018. The technological reality on which the framework was based was the cookie-and-e-commerce reality of around 2014. By 2026, the relevant data flows have shifted to real-time bidding systems, ambient surveillance via the internet of things, biometric recognition in public space and large-scale third-party data trading. The GDPR works on these new categories through interpretation, jurisprudence and guidelines from the European supervisory authorities, while the practices they affect are rolled out in a commercial environment that recalibrates every six months. The Dutch Data Protection Authority publishes annual reports in 2026 that for the most part deal with interpretation of an eight-year-old framework on new cases, and to a much smaller part with anticipatory framework work. Not for lack of competence, but because the legal and institutional instruments are not equipped for anticipation.

Algorithmic equality before the law. The prohibition of discrimination has applied, since the case law on the benefits affair, also to algorithms, provided that the operation and outcomes are transparent to the assessing body. What happens in practice is that the assessing body usually only comes into the picture after the harm has been suffered, not before the model has gone into production. An ex ante systematic test for equality before the law does not exist in the Netherlands as an institutionalised framework. The Human Rights and Algorithms Impact Assessment was revised in February 2026 to Version 2 and aligned with Article 27 of the European AI Act, which from 2 August 2026 makes a fundamental rights impact assessment (FRIA) mandatory for public bodies using high-risk AI systems. IAMA, however, remains an administrative instrument; a stand-alone statutory obligation to conduct an IAMA does not exist as of April 2026, even four years after the Bouchallikh/Dekker-Abdulaziz parliamentary motion intended to enforce it.¹⁴ The Implementation Act for the AI Regulation, which designates the Data Protection Authority and the Dutch Authority for Digital Infrastructure jointly as coordinating national supervisors and gives ten sectoral supervisors competences in addition, entered public consultation on 20 April 2026. The European AI Act sets high-risk requirements from 2 August 2026, provided the harmonised standards are ready in time, which, given the Digital Omnibus proposal of November 2025, is not certain.¹⁵ In the interim vacuum, models go into production without ex ante systematic equality testing, and correction afterwards happens at a scale that is by definition reactive.

Data sovereignty. The Court of Audit published on 15 January 2025 Het Rijk in de cloud: Donkere wolken pakken samen (The State in the Cloud: Dark Clouds Are Gathering), an audit whose figures cannot be other than uncomfortable. The central government uses 1,588 reported cloud services. Of these, seven hundred (forty-four per cent) are public cloud, four hundred seventy-seven private cloud, and for twenty-six per cent it is not known which form is involved. Of one hundred twenty-six services designated as materially public cloud, eighty-four (sixty-seven per cent) had no risk assessment carried out, even though the 2022 government-wide cloud policy makes this mandatory. More than half of the materially public cloud services are procured from Microsoft, Amazon and Google.¹⁶ Court member Ewout Irrgang formulated the sovereignty risk in the accompanying statement in its simplest form: there is a risk that foreign governments, particularly the United States via the CLOUD Act, can view and possibly even alter information of the Dutch central government or of citizens. What the Court of Audit could not provide is a total figure for annual central-government spending on hyperscalers. That figure is missing from every public source and is a verifiable gap in the sovereignty debate.

As of February 2026, the portfolio in the Jetten cabinet is held by State Secretary Willemijn Aerdts (Digital Economy and Sovereignty, Ministry of Economic Affairs). The Rathenau Institute published in July 2025 a triptych From Digital Dependency to Digital Autonomy that unambiguously formulated that society risks losing control, through its digital dependency on a few large technology companies, over core processes in education, healthcare and public service delivery. The WRR factsheet Digital Strategic Autonomy of 29 April 2025 aligned with this. An AIV report of 17 June 2025 drew the geopolitical line. The Cyber Security Council advocated in its advice to the informateur 2025/2026 a structural increase of cybersecurity spending by six hundred ninety million euros per year, in line with the NATO norm of one and a half per cent. Among all these institutions there is substantive convergence in 2026. The actual entrenchment with hyperscalers is deepening in the same period.¹⁷

In all three domains, the same pattern holds. The exogenous speed of technological development is a given that no Dutch intervention can influence. The institutional speed at which the given is addressed, by contrast, is a Dutch choice, made in an architecture whose building blocks largely date from the cycles of the industrial revolution. Between those two speeds lies a chasm that widens every year. What needs to come over from the other side of the chasm cannot be designed by methods from the original architecture.

§ 06 Not incompetence but temporal asymmetry

The diagnosis must be precise. The Dutch public administration is not incompetent in its handling of technology. At the individual level, the central government has highly qualified data analysts, security officers and ethicists. The Centre for Information Security and Privacy Protection, the ICT Assessment Advisory Board, the Rathenau Institute, the Data Protection Authority and the AI team at the Ministry of the Interior deliver work that, in international perspective, hardly lags behind what is done in comparable countries. At the university level too, the Netherlands is strongly represented in algorithmic accountability, AI ethics and technology law.

What is missing is not knowledge. What is missing is an institutional design that lands the available knowledge in working form on the cycles of exogenous reality. The current design assumes that a ministry identifies an issue, requests advice, drafts a bill, puts it out for consultation, considers it, adopts it, brings it into force and implements it. The time between identification and implementation in this design is rarely shorter than five years, often seven to ten. In a world in which a fundamentally new model architecture can move in twelve months from publication to production-wide use, this design rhythm is functionally obsolete. The expertise is there; the wiring in which that expertise lands at operational speed is not.

Whoever turns this into “incompetence” chooses a diagnosis that is consoling for the apparatus because it suggests solutions the apparatus already knows. Better training of ministers, harder selection of senior civil servants, more robust ICT policy. These are measures that stay within the current institutional pace and therefore cannot remedy it. The temporal asymmetry diagnosis is more uncomfortable because it points to a solution that requires the architecture itself to be addressed: a redesign of the wiring, not an upgrade of the staffing.

§ 07 What states do that treat the speed differential as a design problem

In the international literature, four references consistently figure as countries that have tried to align institutional speed with exogenous speed. None of them has solved the problem. All four show how the design principle differs from the Dutch one, and what price they pay for their chosen design. In the OECD Digital Government Index 2023, the Netherlands scores zero point five six, below the OECD average of zero point six one, and in a different light from the three states discussed in this section as references.¹⁸

Denmark — the federal reference model

Denmark is, for the Dutch question, the administratively most comparable case. It is a decentralised unitary state with strong municipalities and regions, a parliamentary democracy, an EU context, and a scale comparable to the Netherlands for critical processes. Since 2001, Denmark has run six successive joint digitalisation strategies, each signed by the state, KL (the Danish municipalities umbrella) and Danske Regioner. The seventh strategy 2026-2029 is in preparation. Since 2022, Digitaliseringsstyrelsen has been operational under a stand-alone Ministry of Digitalisation. The MitID digital identity is used by ninety-six point six per cent of the population over fifteen. In the United Nations E-Government Survey 2024, Denmark ranks first out of one hundred ninety-two countries.¹⁹

The Danish design principle is the pivot here: a central mandate (Digitaliseringsstyrelsen) works with strong local autonomy, because KL and Danske Regioner are co-signatories of the strategy cycles. What the Netherlands could in theory reproduce in its central-government / municipalities / provinces relationship, Denmark has placed in a more binding architecture: joint financing, joint standards, joint priority list per cycle. The price of the Danish design is a time investment at the front end of every cycle, in negotiation among three layers of government that each must put their signature to it. The gain is that decisions, once made, work through to all layers without each layer having to redo its own design work. For the Dutch default, in which each layer of government decides for itself and coherence is left to ad hoc coordination meetings, Denmark is the sharpest illustration that federal interoperability comes about by design and not as a by-product of goodwill.

Estonia — radical centralism as a post-1991 experiment

Estonia has digitalised its central government since 2000 in an architecture in which services address each other via a standardised platform (X-Road) and in which every citizen has a digital identity card that regulates access. The Riigi Infosüsteemi Amet (RIA) operates as the operational core with a statutory monopoly on X-Road connection and RIHA registration, and since 1 January 2024 falls under a combined Ministry of Justice and Digital Affairs.²⁰ The AI and Data Action Programme “Kratt” 2024-2026 has a budget of eighty-five million euros, of which twenty million for open-source AI building blocks. In the UN E-Government Survey 2024 Estonia ranks second; in the OECD DGI 2023, sixth.

The price of the Estonian design is conceptually and contextually heavy. Conceptually: the Estonian single-point-of-accountability model shifts the apparatus’s centre of gravity to one central agency, with a strong statutory mandate. Contextually: this was built in an institutional vacuum after 1991, on a population of just over a million people, in a country that is one of the most unequal societies in Europe. Kattel and Mergel described this in their analysis from 2019 as mission mystique and the hiding hand: the Estonian digital miracle works partly thanks to hidden costs that, in comparable countries, cannot be made invisible.²¹ What remains as a design principle is that institutional speed is a design choice and not a given, and that the choice for a single point of accountability is genuinely a choice, not a self-evidence.

Singapore — centralist speed in a non-democratic context

Singapore launched the Smart Nation programme in 2014, in which a central Government Technology Agency (GovTech) manages the construction, integration and further development of government IT as a single point of accountability. GovTech runs the IT functions for sixty per cent of government agencies through seconded staff. The Singpass digital identity is used by virtually the entire adult population. The Model AI Governance Framework version 2 (2020) was supplemented in 2024 with a Generative AI module and in 2026 with an Agentic AI module. Instruction Manual 8 is the mandatory ICT policy framework for all agencies, in combination with GeBIZ as the central e-procurement portal and the AI Verify toolkit as a sandbox.²²

The governance price is high. Singapore ranks one hundred twenty-sixth out of one hundred eighty on the Reporters Without Borders World Press Freedom Index 2024. SPH Media Trust received three hundred twenty million Singapore dollars in state support in 2022-2023. There is no constitutional right to privacy, and surveillance does not require prior judicial authorisation. The technocratic decision-making that gives Singapore its institutional speed is institutionally entrenched in a way that is not reproducible within an EU-style rule-of-law context. What remains as a design principle is that a single point of accountability is essential for matching institutional and technological time, and that the mandate over procurement is a cornerstone of that match. What is unusable is the governance architecture in which it is embedded.

United Kingdom — illustration of procurement mandate

The United Kingdom is relevant in this section only for one design principle: the mandate over procurement. The Government Digital Service (GDS) has been operational since 2011; the Central Digital and Data Office (CDDO) has had, since 2021, spend controls delegated by HM Treasury, requiring GDS approval of digital expenditure above certain thresholds before departments can commit. In January 2025, CDDO, GDS, the Incubator for Artificial Intelligence and the Geospatial Commission were merged under the Department for Science, Innovation and Technology (DSIT). The AI Safety Institute was renamed the AI Security Institute in February 2025 and given a stronger cyber/security focus.²³

What this means for the Netherlands does not lie in the unitary British state structure (which is not comparable), but in the concrete instrument of spend controls. A delegated power from the Ministry of Finance to a digital service to test expenditure above a threshold ex ante is technically conceivable in the Dutch central government and not institutionally unusual (for IT procurement at sub-national governments something similar applies via the ICT Assessment Advisory Board, but advisory and not binding). What the British example adds is that such an instrument can be binding and, in that way, can influence the pace of the supplier architecture through the gate of central-government procurement.

Switzerland — the price of veto power and consensus

Switzerland has, in this version of the paper, deliberately been moved from main reference to counter-indication. Swiss digitalisation works since 2022 through Digitale Verwaltung Schweiz (DVS), a tripartite organisation of the federation, cantons and municipalities with joint financing (two-thirds federal, one-third cantonal). The Bundeskanzlei sector Digital Transformation and ICT Steering, with about eighty staff, coordinates at federal level. eCH-Verein delivers standards with recommendation status. None of these institutions has cross-level executive authority. Switzerland ranks twenty-sixth in the UN E-Government Survey 2024 and is not assessed in the OECD Digital Government Index 2023 and 2025 due to “lack of relevant data”.²⁴

The Swiss e-ID trajectory is illustrative for the Dutch debate because it shows what the price of consensus and veto power can concretely be. A first e-ID law was rejected in a referendum in 2021, with sixty-four point four per cent against. A new proposal was adopted on 20 December 2024 by the Federal Assembly, and in a second referendum on 28 September 2025 was accepted by a margin of fifty point three nine per cent. The launch of the accompanying wallet app Swiyu is at the earliest in the third quarter of 2026. A total of five years between first proposal and effective acceptance, for one basic building block of digital service delivery. The Bundeskanzlei itself formulates in its strategy documents that digitalisation in Switzerland is, in international comparison, advancing rather slowly.

For the Dutch debate, this is relevant because the Netherlands, in its polder architecture, has a comparable consensus-bound character, without the direct-democratic veto role of Swiss referendums. Whoever advocates a Swiss design for Dutch digitalisation accepts implicitly that decision-making takes place at a pace that structurally lags the exogenous technological cycle. That is not a Statecraft position. What remains usable from the Swiss example is a warning: institutional adaptive capacity that comes about only through full agreement at all levels structurally produces a backlog in a race against technological cycles.

Comparative table

The table below sets the three design principles (single point of accountability, mandate over procurement, anticipatory regulatory space) against their institutional costs by country. The plus and minus signs do not denote a value judgement, but the relative degree to which the principle is institutionally anchored in that country. The Dutch default is included for reference.

Country	Single point of accountability	Mandate over procurement	Anticipatory regulatory space	Institutional costs
Denmark	Strong (Digitaliseringsstyrelsen)	Medium (joint financing)	Medium (cycle strategy)	Front-loaded negotiation among three layers of government
Estonia	Very strong (RIA, X-Road monopoly)	Medium (financial steering)	Medium (sectoral ADM laws)	Path dependency on post-1991 empty state; scale not transferable
Singapore	Very strong (GovTech)	Strong (IM8, GeBIZ)	Strong (Model AI Framework + sandbox)	Non-democratic governance; no constitutional privacy
United Kingdom	Medium (DSIT/GDS-CDDO)	Strong (CDDO spend controls)	Medium (sectoral)	Unitary state structure; post-Brexit context
Switzerland	Weak (DVS consensus)	Soft (AGB standards)	Weak (referendum cycle)	Veto power produces structural backlog
Netherlands (default)	Very weak (fragmented)	Soft (advisory, not binding)	Weak (legislative cycle 5-10 years)	Coordination burden on every layer anew

The Dutch default scores lower on all three design principles than every comparison case, with the result that the institutional costs in the Netherlands manifest in a different place: not in front-loaded negotiation or in curtailment of democratic counter-power, but in cumulative coordination burden on every level that touches on a digital issue. The more rational choice, in line with the Statecraft position, is a mixed form in which federal interoperability (Denmark), procurement mandate (United Kingdom) and anticipatory regulatory space (Singapore, allowing for governance context) are combined under a Dutch administrative arrangement in which the three layers of government are formally co-owners of the digital infrastructure.

§ 08 Handling perspective

The handles that follow from this diagnosis are three.

An institutional place where exogenous speed can land in timely steering. The Dutch apparatus has many advisory bodies, supervisors and knowledge institutes, but no switching point where exogenous technological development is translated, at speed, into institutional steering. No Singaporean GovTech, no Estonian X-Road design team, no Danish Digitaliseringsstyrelsen. Such a switching point need not be large. It must, however, have executive authority over the procurement policy of sub-national governments for the algorithm and AI layer, over the implementation of interoperable standards used by the Dutch executive landscape as a whole, and over a joint strategy cycle co-signed by the central government, IPO and VNG in the Danish manner. The State Secretary for Digital Economy and Sovereignty could fill this role, provided that the position is given an operational service with the same steering relationship as the Inspectorate of Justice and Security has for enforcement, and provided that the procurement mandate, on the British model, is made binding for digital expenditure above a threshold to be determined.

Anticipatory regulation that can keep up with the technological cycle. The legislative cycle, in its current form, is too slow for the digital developments it must address. A complementary framework is needed in which the regulator can, on subdomains and under clearly delineated conditions, set rules that come into force within weeks and are ratified by parliament afterwards. The existing model of orders in council and ministerial regulations offers some footholds. A modified form, with faster ratification and shorter throughput, would do for digital law enforcement what the Emergency Powers Act does for crisis management. Not to disable legislation, but to make its pace fit what it governs. The setting up of regulatory sandboxes on the Singaporean model, with an ethical and human rights anchor that respects the Dutch legal system, is an intermediate form within this that can already be tried out in subdomains.

A sovereign technological infrastructure on European scale, with a Dutch voice in governance. Dependence on American cloud providers for essential executive flows is, in the medium term, not tenable without loss of sovereignty on data issues. A redesign requires a European cloud infrastructure at the level of GAIA-X, but as operational reality and not as coalition proposal. The Netherlands has a role here as a governance voice and as a buyer, not as a unilateral builder. The Court of Audit has published on this. The Rathenau Institute has published on this. The WRR has published on this. The AIV has published on this. The Cyber Security Council has published on this. What is lacking is the administrative translation, and the shift in spending authority needed to make purchasing commitments to a European alternative real.

None of these three handles is unique to the Statecraft diagnosis. They have been advocated by various advisory bodies, researchers and policy officials for some years. What the diagnosis adds is the observation that they should not be treated as ICT issues, but as institutional speed issues. Whoever addresses them as ICT issues stays within the cycles of the apparatus that produces the problem itself. Whoever addresses them as institutional speed issues touches on the architecture that, throughout this Statecraft series, has been identified as the common condition of the diagnosed doorwerkingen.

§ 09 What this means for the Statecraft series

Paper 6, Blind to a Known Future, described the first temporal signature that runs through all five preceding doorwerkingen: systematic failure in the face of futures the apparatus has long known. This paper describes the complementary temporal signature: systematic lagging behind presents that the apparatus cannot keep up with. The two signatures are each other’s opposite, and together they form a vice in which Dutch executive government finds itself trapped in 2026. On the one hand, events whose arrival has been on the forecast for fifteen years and to which the apparatus does not adapt because its steering horizon is shorter than the planning horizon of those events. On the other hand, developments whose arrival is so fast that the apparatus cannot keep up, and which then land in its inner workings without institutional processing. The five form papers have each shown how the vice plays out in a specific domain. Ownership structure, the legal position of the individual citizen, cost shifting between budget holders, social fabric, physical space. In each of those domains, on closer examination, not only the doorwerking of institutional dissociation is visible, but also a specific temporal signature that ensures that the doorwerking is not lifted by the existing institutional cycle.

The closing paper, The Frozen Zeitgeist, brings form and signature together at the level of society. The thesis there will be that the cumulative outcome of five doorwerkingen and two signatures is a frozen spirit of the times in which society has learned to distrust its collective capacity for action. Not because it no longer has that capacity, but because administrative movement itself, in its experience, tends towards failure. The diagnosis of the two signatures explains why this is not a prejudice of an embittered citizen but a rational inference from a twenty-year run of experience.

In the practice layer of the Handbook De Richting van de Beweging (The Direction of the Movement, manuscript in preparation, 2026), the mirror question of this paper is whether the individual interim manager can operate in an environment that knows temporal asymmetry as its baseline.²⁵ The Handbook’s answer is that this is possible, provided the interim cycle is used in such a way that the interim manager’s own working period does not get absorbed in the exogenous cycle but turns it into an institutional adaptation space. For the Statecraft layer at which this paper operates, that means the individual adaptation is no substitute for the systemic redesign that exchanges exogenous speed for institutional steering. Both are necessary. Neither is sufficient.

§ 10 Closing

In April 2026, somewhere in a ministerial decision tree, there is an agenda item about the Algorithm Register Act, with under it the explanation that implementation is being awaited in conjunction with the European AI Act, that the European AI Act enters into force on 2 August 2026, that the accompanying harmonised standards will not be ready in time, that the Digital Omnibus proposal of November 2025 therefore shifts the effective date, that the Implementation Act for the AI Regulation has been in public consultation since 20 April 2026, that IAMA Version 2 has been available since 16 February 2026 but still has no statutory anchoring, and that the House of Representatives passed a motion four years ago whose implementation is still in preparation. Whoever reads the agenda item and sees an individual failure misses the pattern. Whoever sees the pattern reads the agenda item as an institutional declaration of time. No one can order the exogenous speed lowered. Everyone can order the institutional speed raised. The latter is not happening, and the absence of that order has been diagnosed in this series as the core of the doorwerking.

The next paper in this series, The Frozen Zeitgeist, brings five forms and two signatures together in a synthesis on the cumulative deposit at the level of society. There the question is what the sum of the temporal asymmetries diagnosed here does to the self-confidence of a country, and how the Statecraft positioning of constructive diagnosis opens space for the kind of systemic redesign that this paper and its predecessors have advocated.

Colophon

Jacob Huibers is an interim manager with over twenty years of experience in the Dutch public sector. He has worked as cluster manager, cluster director and quartermaster at municipalities ranging from fifty thousand to over two hundred thousand inhabitants and at regional cooperation arrangements in the social and physical domain. Statecraft is his platform for strategic reflection on public execution.

This paper is the seventh in the Statecraft series Reverberation, an eight-part diagnosis of the external symptomatology of institutional dissociation. Lagging Lagging Behind the Speed is the second of two signature essays, alongside Blind to a Known Future (April 2026), which together name the temporal signature that runs through the five form papers. The closing paper, The Frozen Zeitgeist, brings form and signature together in a synthesis on the cumulative deposit at the level of society.

The series builds on the corpus that, in De Richting van de Beweging: Interim-Management in de Publieke Sector (manuscript in preparation, 2026), provides the practice layer; in Allemaal Ontheemd the human layer; and in Decline and Revival the civilisational time layer. The Limbic Literacy work on which § 04 of this paper draws develops the neurological counterpart of the attention asymmetry worked out here in its institutional variant. The Ambtenaar.AI pillar within House of Viridian develops the operational switching point identified in § 08 as the handling perspective.

Contact and subscription: Statecraft.nl/en/contact

Footnotes

¹ Algoritmes.overheid.nl, dashboard and release notes, accessed March 2026: 1,392 algorithm descriptions, 525 participating government organisations. For growth over time: AG Connect, Hoogrisico-algoritmes Justitie komen pas over twee jaar in algoritmeregister (20 December 2023): 264 algorithms by end-2023; iBestuur, Kamer wil dat overheid de transparantie van algoritmes snel verbetert (3 February 2025): 610 in early December 2024, 730 by end-January 2025. See also digitaleoverheid.nl, Succes Algoritmeregister mede dankzij Amsterdam (April 2026), with the milestone of more than 1,300 publications.

² Open State Foundation, op-ed De Volkskrant, January 2025: estimate of sixteen thousand algorithms in active use by Dutch government and approximately two per cent coverage of impactful systems in the Algorithm Register. Compare WRR, Opgave AI: De nieuwe systeemtechnologie, report no. 105 (2021).

³ Dutch Data Protection Authority (Autoriteit Persoonsgegevens), Report on AI & Algorithms in the Netherlands (RAN) 6, March 2026, autoriteitpersoonsgegevens.nl. AI Impact Barometer: four out of nine indicators on red (was two of nine). Three main points: AI in recruitment and selection growing with major risks, transparency and explainability falling short, AI Act preparation lagging. For the DCA position see autoriteitpersoonsgegevens.nl, Department for the Coordination of Algorithmic Oversight (DCA); for capacity question the AP memo A Future-proof AP (Wolfsen, 27 November 2023): seventy FTE on top of existing capacity needed for AI Act, DSA, DMA and Data Act.

⁴ The childcare benefits affair is widely documented. For the algorithmic dimension, see Dutch Data Protection Authority, De verwerking van de nationaliteit van aanvragers van kinderopvangtoeslag (July 2020); ABRvS 14 October 2020, ECLI:NL:RVS:2020:2447; and the findings of the parliamentary committee in Ongekend onrecht (December 2020).

⁵ Parliamentary committee on the childcare benefits affair, Ongekend onrecht, Parliamentary Papers II 2020/21, 35510, no. 2, 17 December 2020. For the algorithmic dimension and its administrative processing: parliamentary inquiry committee on Fraud Policy and Service Delivery, final report Blind voor mens en recht, Parliamentary Papers II 2023/24, 35867, no. 6, presented 26 February 2024 (chaired by M. van Nispen), in particular the parts on risk selection and automation. Cabinet response 5 December 2024, Parliamentary Papers II 2024/25, 35867, no. 13.

⁶ Statecraft position paper De architectuur van het zwijgen: Waarom de datakluis van de Belastingdienst geen incident maar een structureel patroon is (April 2026). For the original announcement of the shielded storage, see the parliamentary letters by State Secretary Snel, April 2019, on the processing of personal data in the context of the Tax Authority’s GDPR-conformity operation.

⁷ Limbic Literacy: How the Attention Economy Hijacks the Brain and What to Do About It, House of Viridian draft manuscript (2026), in particular chapters 4 and 13 on the neurological asymmetry between the limbic system and the prefrontal cortex.

⁸ Van Dam, Pleijers, Hulsegge & Van den Heuvel (2025), Rapport NEA 2024: Resultaten in vogelvlucht, TNO/CBS, May 2025, publications.tno.nl. Burnout complaints 13% (2015) → 20% (2024), average screen time 4.4 hours/day, 44% prolonged screen time. For the trend analysis 2007-2023: Houtman, I., Thielecke, J., De Vroome, E. & Wiezer, N. (2025), Trends in burn-outklachten in Nederland op basis van de Nationale Enquête Arbeidsomstandigheden (NEA), Gedrag & Organisatie 38(2): 109-138, doi:10.5117/GO2025.2.002.HOUT, with the finding that prolonged screen work is the second most strongly explanatory work factor for the rise in burnout complaints over 2007-2023, after work pace. Sickness cost figures via Arbobalans 2024 (TNO 2025-R11108), June 2025: continued-pay costs of work-related sickness €5.1 bn (2015) → €8.3 bn (2023), of which €4.9 bn from psychosocial workload; average sickness duration in burnout 63 days.

⁹ CBS Statline, table 80072NED Ziekteverzuimpercentage; bedrijfstakken (SBI 2008) en bedrijfsgrootte, accessed February 2026; CBS, Ziekteverzuim naar bedrijfstak: ontwikkelingen en verschillen (long read 2024). Public administration Q1 2025: 6.9%; national average 5.2% (2024). ArboNed (2025), Stijgend verzuim door stress alsmaar groeiend probleem voor werkend Nederland, 20 March 2025, arboned.nl: public administration, healthcare and education form top 3 sectors for stress-related absenteeism; average 250 days off due to stress, 318 days due to burnout; nearly a quarter of 2024 absenteeism attributable to long-term stress-related complaints (+7% y-o-y; +35% over five years).

¹⁰ A&O fonds Gemeenten (2025), Personeelsmonitor Gemeenten 2024, May 2025, aeno.nl. Establishment 200,750 employees (+4.6%), absenteeism 6.7% (2024), G4 8.0%, external hire 17.5% of payroll, vacancy rate in municipal sector remains high despite slight loosening of national labour market. Workload perceptions via aeno.nl/verzuim-actueel: 90% of municipalities call workload a current HR issue (stable 2022-2024); 72% took active workload measures in 2024 (was 67% in 2023, 56% in 2022); occupational physicians cite workload and stress in 63% of long-term absence cases. For the 2024 Work Survey (BZK/ICTU/Venster, n=36,522): kennisvandeoverheid.nl/organisatiecultuur/werkonderzoek.

¹¹ Mark, G. (2023), Attention Span: A Groundbreaking Way to Restore Balance, Happiness and Productivity, Hanover Square Press; average focus duration on one screen: 150 sec (2004) → 75 sec (2012) → average 47 sec, median 40 sec (recent measurements, 5 studies); refocus time after interruption: average 25 minutes. For attention residue: Leroy, S. (2009), Why is it so hard to do my work? The challenge of attention residue when switching between work tasks, Organizational Behavior and Human Decision Processes 109(2): 168-181. For the technostress model: Tarafdar, M., Tu, Q., Ragu-Nathan, B.S. & Ragu-Nathan, T.S. (2007), The impact of technostress on role stress and productivity, Journal of Management Information Systems 24(1): 301-328; applied in TNO context in TNO report 2021-R10998 Technostress (Inspectorate SZW, 2021) and the TNO brochure Technostress reikt verder dan alleen technologie (2023).

¹² Microsoft (2025), Breaking Down the Infinite Workday, Work Trend Index Special Report, June 2025, microsoft.com/en-us/worklab/work-trend-index/breaking-down-infinite-workday; n=31,000 knowledge workers in 31 markets including the Netherlands, Edelman Data x Intelligence February-March 2025. Microsoft 365 users are interrupted on average every 2 minutes; average 153 Teams messages and 117 emails per day. Industry research with methodological limitations, used here as illustrative anchor, not as academic source.

¹³ Bovens, M. & Zouridis, S. (2002), From street-level to system-level bureaucracies: How information and communication technology is transforming administrative discretion and constitutional control, Public Administration Review 62(2): 174-183. Soares, C., Grimmelikhuijsen, S. & Meijer, A. (2024), Screen-level bureaucrats in the age of algorithms: An ethnographic study of algorithmically supported public service workers in the Netherlands Police, Information Polity 29(3): 277-292, doi:10.3233/IP-220070. For Dutch public execution more broadly: Borst, R. (2024), Radboud doctoral thesis on digitalisation and public values at the Tax Authority, SVB and UWV. For the Work Survey figure: ICTU/Venster (2024).

¹⁴ Human Rights and Algorithms Impact Assessment Version 2, published 16 February 2026, rijksoverheid.nl/documenten/rapporten/2026/02/16/impact-assessment-mensenrechten-en-algoritmes; aligned with Article 27 AI Act (FRIA), marked for high-risk AI systems. The Bouchallikh/Dekker-Abdulaziz parliamentary motion (5 April 2022) on statutory IAMA was adopted, but the administrative obligation via the Implementation Framework Algorithms (IKA) remains, as of April 2026, the only route. For the Algorithm Register self-classification, see Algorithm Audit / Binnenlands Bestuur (autumn 2025): 14 ministerial letters with collectively about 250 impactful algorithms, of which at the Ministry of Finance 162 impactful but zero high-risk per the department.

¹⁵ European Commission, AI Act, Regulation (EU) 2024/1689; AI Regulation Implementation Timeline, with the relevant dates 2 February 2025 (prohibited practices and AI literacy), 2 August 2025 (governance and GPAI), 2 August 2026 (high-risk and general application) and 2 August 2027 (high-risk embedded in regulated products). For the postponement proposal see the Commission’s Digital Omnibus package (19 November 2025). For the Dutch implementation: public consultation Implementation Act for the AI Regulation (UAIA), 20 April - 1 June 2026, nederlanddigitaal.nl/actueel/nieuws/2026/04/21/internetconsultatie-uitvoeringswet-ai-act-van-start; designation of the Data Protection Authority and Dutch Authority for Digital Infrastructure as joint coordinating national supervisors, plus ten sectoral supervisors. Maximum fines under the AI Act: €35 million or 7% of global turnover for prohibited AI practices.

¹⁶ Court of Audit (2025), Het Rijk in de cloud: Donkere wolken pakken samen, adopted 19 December 2024, made public 15 January 2025, rekenkamer.nl/publicaties/rapporten/2025/01/15/het-rijk-in-de-cloud. 1,588 reported cloud services; 700 (44%) public cloud, 477 private cloud, 26% type unknown. Of 126 services classified as materially public cloud, 67% (84 services) had no risk assessment carried out as required by the 2022 government-wide cloud policy. Quote: “More than half of the materially public cloud services are procured from Microsoft, Amazon and Google.” Irrgang quote via accompanying Court of Audit statement, 15 January 2025. The report contains no total figure for annual central-government spending on hyperscalers; this figure is not available from any primary public source. Additional: Aanhangsel TK 2024-2025, no. 1597 (SSC-ICT migration of 58,000 workplaces).

¹⁷ Rathenau Institute (2025), triptych Van digitale afhankelijkheid naar digitale autonomie, 9 July 2025, rathenau.nl. WRR (2025), Factsheet Digitale strategische autonomie (Prins & Sheikh), 29 April 2025, at the request of the standing parliamentary committee on Digital Affairs, wrr.nl. AIV (2025), AI: technology, power, and democratic values in the foreign policy of the Netherlands, 17 June 2025. Cyber Security Council, advice to informateur 2025/2026: structurally €690 million per year additional cybersecurity spending, in line with NATO norm of 1.5%. State Secretary Willemijn Aerdts (Digital Economy and Sovereignty, Ministry of Economic Affairs) since 23 February 2026 in the Jetten cabinet.

¹⁸ OECD (2024), 2023 OECD Digital Government Index, OECD Public Governance Policy Papers no. 44, doi:10.1787/1a89ed5e-en. Netherlands: 0.56, below the OECD average of 0.61. Top 10 (2023): Korea, Denmark, UK, Norway, Australia, Estonia, Colombia, Ireland, France, Canada. In the now-published OECD DGI 2025 (OECD, February 2026, doi:10.1787/dbe102ed-en) the top three are Korea, Australia and Portugal, with the OECD average shifted upward to 0.70. The Dutch score runs in that measure from 0.32 (2023) to 0.51 (2025). The Netherlands therefore remains below the OECD average, even after revision.

¹⁹ Digitaliseringsstyrelsen (2025), 25 års fælles digitaliseringsstrategier, digst.dk; The Joint Government Digital Strategy 2022-2025 (sixth consecutive since 2001), signed by the state, KL and Danske Regioner; successor 2026-2029 in preparation. MitID adoption 96.6% of population over 15 (Queue-it, How Denmark Became a Digital Government Global Leader, 2024). United Nations (2024), E-Government Survey 2024, EGDI 0.9847 (rank 1 of 192).

²⁰ Riigi Infosüsteemi Amet (RIA), ria.ee/en/authority-news-and-contact/authority-and-management/tasks-and-structure-authority. Since 1 January 2024 under the Ministry of Justice and Digital Affairs. Kratt 2024-2026 (AI & Data Action Plan), €85 mln budget, €20 mln for open-source AI building blocks; interoperable-europe.ec.europa.eu/collection/open-source-observatory-osor/news/estonia-invests-eu20-million-open-source-ai. UN E-Government Survey 2024 (rank 2, EGDI 0.9727); OECD DGI 2023 (rank 6, score 0.74). For procurement figures OECD (2025), Strategic Public Procurement and Professionalisation Initiatives in Estonia: Full Report, doi:10.1787/56e03f80-en.

²¹ Kattel, R. & Mergel, I. (2019), Estonia’s Digital Transformation: Mission Mystique and the Hiding Hand, in: M. Compton & P. ‘t Hart (eds.), Great Policy Successes, Oxford University Press, doi:10.1093/oso/9780198843719.003.0008. Contains the analysis of path dependency on post-Soviet “empty state” 1991, social cost (Estonia is one of the most unequal societies in Europe) and scale incompatibility (1.375 mln inhabitants versus NL 17.9 mln).

²² GovTech Singapore, tech.gov.sg/about-us/who-we-are/; Singapore Smart Nation 2 (1 October 2024), mddi.gov.sg/newsroom/smart-nation-2-press-release/. UN MSQ2024 Singapore submission, publicadministration.un.org: GovTech runs IT functions for 60% of government agencies via seconded staff. Model AI Governance Framework v2 (2020); Generative AI version (May 2024); Agentic AI version (January 2026). Reporters Without Borders, World Press Freedom Index 2024: Singapore 126 of 180. UN E-Government Survey 2024 (rank 3, EGDI 0.9691).

²³ Government Digital Service, gov.uk; CDDO (Central Digital and Data Office) and GDS merged under DSIT, January 2025, publictechnology.net/2025/01/21/education-and-skills/cddo-brought-back-into-gds-in-digital-government-shake-up/; spend controls delegated by HM Treasury, gov.uk/government/publications/roadmap-for-digital-and-data-2022-to-2025. AI Safety Institute renamed AI Security Institute as of 14 February 2025, freevacy.com/news/uk-government/uk-announces-ai-safety-institute-will-become-ai-security-institute/6135. UN E-Government Survey 2024 (rank 7); OECD DGI 2025 (rank 4, score 0.84).

²⁴ Bundeskanzlei, Digitale Transformation und IKT-Lenkung, bk.admin.ch/bk/en/home/digitale-transformation-ikt-lenkung/bereichdti.html; Digitale Verwaltung Schweiz (DVS), digitale-verwaltung-schweiz.ch, since 1 January 2022; DVS Strategy 2024-2027 (effective 1 January 2024). UN E-Government Survey 2024 (rank 26); not assessed in OECD DGI 2023 and 2025 due to “lack of relevant data”. First e-ID law rejected in 2021 referendum (64.4% against); new proposal adopted by Federal Assembly 20 December 2024; second referendum adopted 28 September 2025 with margin of 50.39%; launch of wallet app Swiyu at earliest Q3 2026. Library of Congress, Switzerland: Voters Approve Electronic ID Law in National Referendum, Global Legal Monitor, 4 December 2025, loc.gov/item/global-legal-monitor/2025-12-04/.

²⁵ J. Huibers, De Richting van de Beweging: Interim-Management in de Publieke Sector (manuscript in preparation, 2026), in particular the chapters on the interim cycle, anchoring as KPI and the Aiki method as institutional design principle.