Change log

1. Version 0.9.0 · 31 October 2026

   A library of what to read

   • A new library at /library gathers 90 laws, papers, books, trackers and courses by other people. Each entry says what it gives you, what it leaves out, how long it takes, and which parts of the framework it speaks to.
   • The library opens with the whole collection drawn as a shelf: one spine per resource, standing in the year it came out, its height the time it takes to read. Four reading paths follow, for people new to AI policy, coming from tech, working in or on Africa, or in government.
   • Every entry can be browsed and filtered at /library/all.
   • The reference shelf of our own framework, sources and data, which used to be called the library, is now at /reference.

2. Version 0.8.0 · 29 October 2026

   A working paper, and a history that reads as one

   • The site now opens as the front page of a working paper: the definition, the argument in brief, three ways in, and a figure of the four objects, the four tests and time. The main menu is Read, History and Use it.
   • The history has been rewritten. It now explains who each body is, follows how the idea of timing appears, fades and returns, and says plainly whose voices the record is missing. All 38 documents can be browsed, filtered and downloaded as a spreadsheet.
   • Use it opens with the eleven questions, and they can be printed on one page.
   • A new page, /paper, says where the research paper stands, what it has already changed, and how to help finish it.
   • The essay opens with the argument in five sentences.

   Corrections

   • The history and the slides said no earlier sentence named what governance acts on or built timing into the definition. A draft review found earlier work that had reached parts of both: Domínguez Hernández and colleagues, UNESCO, Kolt and colleagues, and Aslam. They are now credited, and the claim is narrowed to the combination in one sentence.
   • A history finding said timing is raised far less often than adaptation. Timing is often discussed in the documents; it is rarely written into their definitions. The finding now says so.
   • Slide 10 repeated the word power in its heading.

3. Version 0.7.0 · 27 October 2026

   Production clean-up

   • The framework page is now a compact reference card; the argument lives in the essay and each part is read closely on the definition page.
   • Object and test pages now explain first and list related records second. Worked-analysis answers are linked rather than reprinted.
   • The three original worked analyses now separate what the scenario stipulates from what follows, and list what remains unknown.
   • This public change log replaces references to a repository file.

   Corrections

   • A test is a question that helps examine an arrangement; an earlier sentence said the tests “decide” whether it is defensible.
   • Incentives: a principle that ignores incentives is likely to lose to them, not certain to (“every time” removed).
   • Leftover wording from the previous definition was updated: “patterns, not incidents” now reads as attention to patterns without excluding individual harm, and the temporal principle is keeping the capacity to change course.
   • Brazil’s Bill 2338/2023 is labelled as a bill, not a law.
   • Ten claims shown with high confidence had no checked passage; their confidence is now capped at medium until a passage is checked.
   • Links to raw metadata records were replaced with the article’s DOI.
   • Links to unpublished drafts were removed; the drafts are described in the framework’s provenance.
   • The corrections page now describes the channel that actually exists.
   • The accountability guide is labelled a draft: it was source-checked but has not been reviewed by the author or an outside reviewer.

4. Version 0.6.0 · 25 October 2026

   Broadsheet redesign and a revised definition

   • The canonical definition was revised by the author, with a matching short form.
   • New pages: the argument, the definition unpacked, how the parts connect, one decision followed through, history, and slides.

5. Version 0.5.0 · 19 October 2026

   Definitions compared and human-scale governance

   • A date-ordered reading of how 38 documents define or frame AI governance.
   • A sixth level of governance: community and organisational use.
   • New worked analyses for a school, a community organisation and a child-safety regulator, with their unknowns listed.

6. Version 0.4.1 · 13 October 2026

   First public research preview

   • Expanded readings of each part of the definition and the four objects, with source trails.

7. Version 0.3.0 · 7 September 2026

   The framework as the backbone

   • The four objects, four tests, lifecycle, levels and eleven questions became the structure of the site.

Found something to correct? How to send a correction. Every version of the dataset is listed in the data index.