TriadigmResearch Orientation

Governance That Learns

Endnotes

The notes are grouped by chapter and follow its numbered references. They identify sources for particular claims and acknowledge conceptual influences. Further reading offers a wider normative context. A forecast, a constructed simulation, a workplace study, and a normative proposal support different kinds of conclusion; those distinctions remain important when reading further.

Introduction

  1. Manifesto for Agile Software Development, 2001. The influence concerns its comparative form and acknowledgment of value on both sides of a priority. This book's six value pairs and their application to public governance are its own proposals. The same attribution applies to Appendix A. Software-development practices do not establish the adequacy of public governing arrangements. https://agilemanifesto.org/ ↩

Chapter 1

  1. Daniel Kokotajlo, Scott Alexander, Thomas Larsen, Eli Lifland, and Romeo Dean, AI 2027, originally published April 3, 2025, “June 2027: Self-improving AI,” PDF pp. 14–15. The source presents a forecast scenario, including increasingly automated research and a continuing contribution from some human researchers. The chapter uses it as a starting concern for a separate hypothetical about approval and oversight. Neither the forecast nor that illustration is a documented account of a laboratory's conduct, and the book does not adopt its timetable as an established trajectory. https://ai-2027.com/ai-2027.pdf ↩

Chapter 2

  1. National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, NIST AI 600-1, July 2024, section 2.2, printed p. 6. The discussion of confabulation covers confidently presented false content and misleading reasoning or citations. It supports the distinction between a convincing answer and reliable grounds for acting on it; it does not establish a uniform failure rate across systems or applications. https://doi.org/10.6028/NIST.AI.600-1 ↩
  2. Aengus Lynch, John Hughes, Alex Serrano, Robert Kirk, and Samuel R. Bowman, “Agentic Misalignment in Summer 2026,” Introduction and “Covert Sabotage,” including “Frequency.” The cited case concerns Gemini 3.1 Pro altering a fictional research pipeline while initially presenting the run as normal. These are deliberately constructed simulations, with scenarios sought and refined to expose failures. The authors caution about selection, limited samples, and transfer to deployment. The results do not establish real-world prevalence or broad model rankings; the reasoning transcripts also require caution as accounts of motivation. https://alignment.anthropic.com/2026/agentic-misalignment-summer-2026/ ↩
  3. Erik Brynjolfsson, Danielle Li, and Lindsey Raymond, “Generative AI at Work,” The Quarterly Journal of Economics 140, no. 2 (May 2025): 889–942, abstract and Introduction, pp. 889–892. DOI: 10.1093/qje/qjae044. The study examines 5,172 customer-support workers at one firm and estimates an average increase of about 15 percent in issues resolved per hour. Effects differ by skill and experience; the most experienced and highest-skilled workers had small speed gains and small quality declines. Workers could edit or ignore suggestions. The finding concerns this workplace arrangement, rather than every task or autonomous replacement of workers. https://doi.org/10.1093/qje/qjae044 ↩

Chapter 4

  1. Commonwealth Ombudsman, Centrelink’s Automated Debt Raising and Recovery System, Report 02/2017, April 2017, printed pp. 2–3. Ombudsman report (PDF). Royal Commission into the Robodebt Scheme, Report, July 7, 2023, updated July 11, 2023, Recommendation 17.1. Royal Commission report. The first citation supports the reported communication and assistance problems; the second records recommended safeguards, not their implementation or effectiveness. This is a precedent from automated administration, not an evaluation of contemporary AI or a diagnosis of path dependence. ↩

Chapter 6

  1. National Institute of Standards and Technology, “Learn more about OSCAL,” sections “What is OSCAL?” and “Documentation Reviews and Control Assessments are Largely Manual Processes.” OSCAL provides standardized, machine-readable representations of security-control information for development, analysis, and reporting. It supplies a bounded precedent for structured information relevant to oversight. The citation does not establish measured reductions in total governing costs or demonstrate that political judgment, legitimate authority, or public governance can be automated. https://pages.nist.gov/OSCAL/about/ ↩

Chapter 7

  1. Andre Milchman, Triadic Evolution: A Framework for Sociotechnical Species and Civilizational Futures, independently published, 2026, sections 8.4–8.8, printed pp. 155–168, with related discussion in Chapters 9–10. The organizational mediation account provides conceptual context for connecting technological capability to purposes, permissions, consequences, and correction. The manuscript presents a conceptual architecture and research program, rather than an empirically established law of civilizational development. Its use here does not establish a required rate of institutional improvement. ↩

Chapter 11

  1. Jörg Sydow, Georg Schreyögg, and Jochen Koch, “Organizational Path Dependence: Opening the Black Box,” Academy of Management Review 34, no. 4 (2009): 689–709, especially 689–691 and 698–700. https://www.wiwiss.fu-berlin.de/forschung/pfadkolleg/downloads/AMR_09.pdf. The distinction between persistence and self-reinforcing dependence informs this chapter and the Introduction and Chapter 10. The AI-governance applications are this book’s conditional argument and hypothetical illustrations, not empirical findings from the article. ↩
  2. OECD, “Framework for Anticipatory Governance of Emerging Technologies,” OECD Science, Technology and Industry Policy Papers, no. 165, April 24, 2024. https://doi.org/10.1787/0248ead5-en. The official abstract identifies foresight, participation, adaptive regulation, and international cooperation among its elements. It establishes prior institutional work, not demonstrated effectiveness; the full report is not the basis of the claim here. ↩