About this site
This page explains how A Plurality Atlas is made and who runs it, so that you can weigh what you read here.
How it is made
- Collecting materials — Centred on the book Plurality (its English, Japanese and Traditional Chinese editions, all CC0) and the 6-Pack of Care (civic.ai), we collected talks, conversations and interviews of 25 people, starting with Audrey Tang and Glen Weyl (about 780 items), news articles, blog posts and commentary in 30 languages (about 1,350 items), papers and more. As of October 2026 there are about 4,000 items. For each one we record the original URL, the date it was retrieved and its license. Because some of them are copyrighted, the collection itself is not public.
- Building a working wiki — The AI read the materials and organised them into pages on concepts, people and chapters of the book (about 180 pages). Every statement carries a note saying which material it comes from. This follows the "LLM Wiki" approach proposed by Andrej Karpathy.
- Writing the articles — From the wiki pages, the AI wrote articles for first-time readers. The rules: nothing that cannot be confirmed in the materials, and if the materials contain criticism or debate, it goes in. Most topics were chosen from the book and the collected materials, but a few were chosen by the operator because he thought they would be useful to readers (Cybozu; Keio University's X Dignity Center and PRANCE). Those articles say so in the text.
- Checking — Articles were checked in two passes: first against the wiki, then directly against the original materials, bypassing the wiki. The second pass found 89 errors in 116 articles (dropped caveats and conditions, "inspired by" turned into "based on", and so on), which were corrected. Since then, another AI has re-read the site as a critical reader, and the problems it found have been fixed.
- Other languages — The English and Traditional Chinese articles were also written by the AI, with terms checked against each language edition of the book. The Simplified Chinese version is an automatic character conversion (OpenCC) of the Traditional Chinese articles, so much of its wording is Taiwanese usage.
- Publishing — The site is published on GitHub Pages. The text is CC0 (public domain). Quoted material belongs to its sources; where the license is known, it is shown in [ ] in each article's Sources list.
The operator and conflicts of interest
The site is run by Hirokazu Nishio (Cybozu Labs). He is personally involved in some of the topics it covers:
- He contributed to editing the English edition of Plurality and led the project to publish the Japanese edition.
- Cybozu, the publisher of the Japanese edition (through its Cybozu Shiki Books imprint), is his employer. Cybozu Labs is Cybozu's research and development unit.
- He provided technical support to Takahiro Anno's 2024 Tokyo gubernatorial campaign.
- He is one of the main developers of Kouchou AI, a Digital Democracy 2030 project.
- He is one of the authors of the Japanese book Broad Listening, co-written by practitioners.
- He volunteers on the technical side of Team Mirai's projects (such as its "talking manifesto"). He is not employed by the party.
Articles about practice in Japan (broad listening, Kouchou AI, Takahiro Anno, Team Mirai, Cybozu and others) rely heavily on materials he wrote. Please keep this involvement in mind when reading them.
Limitations
- Because the articles were written by AI, errors may remain. What was checked is whether they match the collected materials; nothing outside those materials was checked.
- The materials are mostly in English, Japanese and Chinese, with fewer in other languages. Counts in the articles on how Plurality has been received in various countries ("as far as this site has collected") are not exhaustive surveys.
- The articles are based on materials available in September–October 2026.
Counting visits
Since 16 November 2026, this site has used GoatCounter, a web analytics service (a tool that counts how often pages are opened), to learn which articles are read, in which countries and in which languages. We use this to decide which articles to add and which to improve.
- What is recorded — The page that was opened, the page the visitor came from, the browser and operating system (at the level of "Firefox 139" or "Windows 11"), the screen width, the language setting and the country. The country is worked out from the IP address (the number that identifies a connection on the internet). The records are totals per hour and per day; individual page views are not recorded.
- What is not recorded — The IP address itself is not stored. Nothing is stored in the visitor's browser: no cookies (small pieces of data a site saves in a browser) or anything similar.
- Repeat visits — So that the same person's visit is not counted again and again, the combination of site, IP address and browser information is replaced with a random string. That mapping is kept only in the memory of GoatCounter's server, for up to eight hours.
- Where the data is and who runs it — GoatCounter is an open-source service (its program code is public) operated by Martin Tournoij, who is located in Ireland. The data is stored on servers of Hetzner Online GmbH in Finland and Germany. GoatCounter states that no information is shared with third parties.
- About consent notices — This site does not show a cookie or consent notice. GoatCounter explains that it "probably" does not require one, on the grounds that, among other things, no personally identifiable information is collected. GoatCounter itself notes that this is not a lawyer's opinion.
- If you do not want to be counted — Open any page of this site with
#toggle-goatcounteradded to the end of the URL, and views from that browser will no longer be counted (a confirmation appears; this one setting is stored in your browser). Doing the same again turns counting back on. You can also block GoatCounter with an ad blocker or similar; the articles remain readable.
GoatCounter's own explanations: https://www.goatcounter.com/help/privacy and https://www.goatcounter.com/help/gdpr
To learn how this site is found through Google Search, we also use Google Search Console (a Google tool that lets a site's owner see the words people searched for, how often the site appeared in search results, and how often it was clicked). It shows figures compiled on Google's side of the search; it does not add anything to these pages that counts visitors.
If you find a mistake
Please let us know through the link at the end of each article or via GitHub issues (https://github.com/nishio/plurality-atlas/issues). Submissions are public. Replies are written by the AI after checking the materials; anything that needs judgment, such as matters touching on politics or on the evaluation of particular people or organisations, is reviewed by Nishio.