修訂紀錄

2026-10

1. 2026-10-13/compassion-and-ai/* as heard
2. 2026-10-13/tw/ · /tw/ai-crisis-diplomacy/ · /tw/ai-democracy-podcast/ · /tw/ciudadania-digital/ · /tw/doom-debate/ · /tw/outrage-to-overlap/ · /tw/plurality-ucsc/ · /tw/virtuous-ai/tw: keep 共創無限未來 for free the future, as Audrey's locked rendering
3. 2026-10-13/6/ · /compassion-and-ai/ · /tw/ · /tw/1/ · /tw/2/ · /tw/3/ · /tw/4/ · /tw/5/ · /tw/6/ · /tw/ai-alignment-cannot-be-top-down/ · /tw/ai-crisis-diplomacy/ · /tw/ai-democracy-podcast/ · /tw/ciudadania-digital/ · /tw/civic-ai-dialogue/ · /tw/democracy-needs-civic-ai/ · /tw/doom-debate/ · /tw/faq/ · /tw/manifesto/ · /tw/measures/ · /tw/outrage-to-overlap/ · /tw/plurality-ucsc/ · /tw/podcast/ · /tw/reimagining-ai-alignment/ · /tw/software-freedom-as-civic-care/ · /tw/sources/ · /tw/virtuous-ai/tw: hold translations to the English (SubQ audit), credit pack 6 authors, fix compassion transcript
4. 2026-10-13/glossary/ · /tw/ · /tw/1/ · /tw/2/ · /tw/3/ · /tw/4/ · /tw/5/ · /tw/6/ · /tw/ai-crisis-diplomacy/ · /tw/ai-democracy-podcast/ · /tw/compassion-and-ai/ · /tw/conference/ · /tw/democracy-needs-civic-ai/ · /tw/doom-debate/ · /tw/faq/ · /tw/glossary/ · /tw/inside-the-kami/ · /tw/manifesto/ · /tw/podcast/ · /tw/reimagining-ai-alignment/ · /tw/safer-sovereignty/ · /tw/software-freedom-as-civic-care/ · /tw/sources/ · /tw/we-the-people-are-truly-the-superintelligence/tw: match the book's revised terms: 熱情女主人, 超級智慧, 惡意 AI 蜂群, 複合案例, 金絲雀發布, 解放未來, 約束力
5. 2026-10-11/ · /tw/home: say the instructions guide and the group holds the AI to them
6. 2026-10-08/sources/ · /tw/sources/spell Ravid Shwartz Ziv without the hyphen, as he does
7. 2026-10-08/care-ai/ · /measures/ · /safer-sovereignty/ · /tw/care-ai/ · /tw/measures/ · /tw/safer-sovereignty/run-in lead terms take a period instead of a dash, as in the book: measures, safer sovereignty, care AI
8. 2026-10-04/ledger/ · /tw/ledger/page the amendment ledger by month so no ledger page outgrows the size budget
9. 2026-10-04/5/ · /tw/1/ · /tw/2/ · /tw/3/ · /tw/4/ · /tw/5/ · /tw/6/pack 5 closes on the book's quilt without a backing; zh closing images are 收尾意象, as in the book
10. 2026-10-04/ · /tw/home: lead with the group staying in charge; the instructions only ask (EN and zh-TW)
11. 2026-10-04/ · /1/ · /2/ · /3/ · /4/ · /5/ · /6/ · /comics/ · /faq/ · /glossary/ · /inside-the-kami/ · /ja/comics/ · /kami/ · /measures/ · /openclaw/ · /tw/ · /tw/1/ · /tw/2/ · /tw/3/ · /tw/4/ · /tw/5/ · /tw/6/ · /tw/comics/ · /tw/faq/ · /tw/glossary/ · /tw/inside-the-kami/ · /tw/kami/ · /tw/measures/ · /tw/openclaw/work in the critical-friend site review: plain first screen, honest Kami, limits in code
12. 2026-10-03/glossary/ · /tw/glossary/glossary: clarify appeal, brake and data-coalition requirements
13. 2026-10-03/6/ · /tw/6/pack 6: distinguish the kami metaphor from institutional design
14. 2026-10-03/inside-the-kami/ · /tw/inside-the-kami/inside-the-kami: credit SAI authors and bound learning-theory claims
15. 2026-10-03/2/ · /sources/ · /tw/2/ · /tw/sources/pack 2, sources: update detected-ad figures and retain evidence limits
16. 2026-10-02/2/ · /3/ · /ai-alignment-cannot-be-top-down/ · /ai-democracy-podcast/ · /ciudadania-digital/ · /democracy-needs-civic-ai/ · /doom-debate/ · /faq/ · /glossary/ · /kami/ · /manifesto/ · /reimagining-ai-alignment/ · /sources/ · /tw/2/ · /tw/3/ · /tw/ai-alignment-cannot-be-top-down/ · /tw/ai-democracy-podcast/ · /tw/ciudadania-digital/ · /tw/democracy-needs-civic-ai/ · /tw/doom-debate/ · /tw/faq/ · /tw/glossary/ · /tw/kami/ · /tw/manifesto/ · /tw/reimagining-ai-alignment/ · /tw/sources/ · /tw/we-the-people-are-truly-the-superintelligence/ · /we-the-people-are-truly-the-superintelligence/carry the book's final corrections to the site (EN and zh-TW in parity)
17. 2026-10-02/faq/ · /sources/ · /tw/faq/ · /tw/sources/faq, sources: anchor the pause discussion in the Statement on Superintelligence, which Audrey signed; Sources lists it (EN and zh-TW in parity; anchor ledger re-recorded)
18. 2026-10-02/ · /2/ · /faq/ · /glossary/ · /inside-the-kami/ · /kami/ · /manifesto/ · /sources/ · /tw/ · /tw/2/ · /tw/faq/ · /tw/glossary/ · /tw/inside-the-kami/ · /tw/kami/ · /tw/manifesto/ · /tw/sources/faq, home, kami, inside-the-kami, pack 2, sources, glossary, manifesto note: stop pointing readers into the book; FAQ Q17 links and describes the Noema assembly proposal (EN and zh-TW in parity; anchor ledger re-recorded)
19. 2026-10-02/faq/ · /tw/faq/faq: describe the call for a global pause without singling out one person or calling it a third position (EN and zh-TW in parity; anchor ledger re-recorded)
20. 2026-10-02/ · /5/ · /faq/ · /manifesto/ · /safer-sovereignty/ · /software-freedom-as-civic-care/ · /sources/ · /tw/ · /tw/5/ · /tw/faq/ · /tw/manifesto/ · /tw/safer-sovereignty/ · /tw/software-freedom-as-civic-care/ · /tw/sources/ · /tw/we-the-people-are-truly-the-superintelligence/ · /we-the-people-are-truly-the-superintelligence/pack 5, home, faq, sources, editorial notes: ROOST Model Community framing, the distribution and practice layer that makes federated training of safety models feasible (EN and zh-TW in parity; anchor ledger re-recorded)
21. 2026-10-02/faq/ · /inside-the-kami/ · /tw/faq/ · /tw/inside-the-kami/faq, inside-the-kami: sync with the book's Krueger changes
22. 2026-10-02/2/ · /5/ · /glossary/ · /sources/ · /tw/2/ · /tw/5/ · /tw/glossary/ · /tw/sources/sync packs 2 and 5, glossary and sources with the book; add Krueger source
23. 2026-10-02/manifesto/ · /tw/manifesto/refactor: just say "precursor"
24. 2026-10-02/ · /1/ · /2/ · /3/ · /4/ · /5/ · /6/ · /faq/ · /glossary/ · /inside-the-kami/ · /kami/ · /manifesto/ · /measures/ · /sources/ · /tw/ · /tw/1/ · /tw/2/ · /tw/3/ · /tw/4/ · /tw/5/ · /tw/6/ · /tw/faq/ · /tw/glossary/ · /tw/inside-the-kami/ · /tw/kami/ · /tw/manifesto/ · /tw/measures/ · /tw/sources/packs, glossary, Kami essays: fix the site-pertinent book-review items
25. 2026-10-02/1/ · /2/ · /glossary/ · /measures/ · /tw/1/ · /tw/2/ · /tw/glossary/ · /tw/measures/packs 1–2, glossary, measures: fix pertinent book-review items
26. 2026-10-02/manifesto/ · /sources/ · /tw/manifesto/ · /tw/sources/manifesto, sources: record X's Community Notes study (arXiv 2609.40067) — a closing editorial note that the speech's two lanes, AI for speed and humans for judgement, have since been measured on X's AI Note Writer API rather than Collaborative Notes (in June 2026 the Grok-based Community Writer drafted 41 per cent of proposed notes and 52 per cent of the notes rated helpful, while every note was still shown only after cross-perspective rating; the platform's own figures), and a Pack 5 Sources entry (EN and zh-TW in parity; anchor ledger re-recorded; parity baseline: the only change is the new p113a on each manifesto page)
27. 2026-10-01/glossary/ · /tw/glossary/gloss: skip everyday uses of glossary words
28. 2026-10-01/sources/ · /tw/sources/sources: correct the CIP section against cip.org — Safety Paradox is Padolsey's, Shared Code is Schneider, Siddarth and Stein, network societies adds Weyl and Slaughter, Predistribution adds Manning, Pulse is Brandon Jackson's guest post, When Safety Evals and Nine Months In are CIP's; Summerfield now cites Nature Human Behaviour 9 (2025) and Hamna CHI 2026; add A Vision of Democratic AI, Burton et al., the Columbia Convening, Luettgau et al., the Global Representativeness Index, Samiksha with Anthropic's Bengaluru account, the CI Corporation brief, the trust snapshot, the annual reports, and Democracy Levels; credit Flynn Devine on the Roadmap (EN and zh-TW in parity)
29. 2026-10-01/comics/ · /glossary/ · /ja/comics/ · /openclaw/ · /tw/comics/ · /tw/glossary/ · /tw/openclaw/comics: vector plates with live, selectable text
30. 2026-10-01/comics/ · /glossary/ · /ja/comics/ · /openclaw/ · /tw/comics/ · /tw/glossary/ · /tw/openclaw/ja comics: drop overlay tab stops, test gallery output
31. 2026-10-01/conference/sensemaking/ · /tw/conference/sensemaking/fix CI: restore the Mandarin Polis question to the wording the sensemaker's pinned evidence and narratives were generated from (「人們願意接受哪一種 AI 應用於照護、政府與日常公共生活？」), so the deterministic replay verifies again; rewording it would require regenerating the pinned narratives
32. 2026-10-01/tw/ai-with-us/ · /tw/care-ai/ · /tw/conference/ · /tw/conference/report/apply the Mandarin read's remaining label fixes at the owner's call: 提供佐證 for 發送收據 in Care AI (checkable evidence, not purchase receipts), 歡迎致詞與為今日議程定調 and 影片：AI 與我們同在，而非撇開我們 in the conference agenda (and the video page's own title), 社群談 AI 的聲音 as the community report's title, and 匿名投票群體 in the Polis report (parity baseline: the two changed page titles)
33. 2026-10-01/comics/ · /conference/sensemaking/ · /glossary/ · /ja/comics/ · /openclaw/ · /tw/ · /tw/1/ · /tw/2/ · /tw/3/ · /tw/4/ · /tw/5/ · /tw/6/ · /tw/ai-alignment-cannot-be-top-down/ · /tw/ai-crisis-diplomacy/ · /tw/ai-democracy-podcast/ · /tw/ai-with-us/ · /tw/care-ai/ · /tw/ciudadania-digital/ · /tw/civic-ai-dialogue/ · /tw/comics/ · /tw/compassion-and-ai/ · /tw/conference/ · /tw/conference/sensemaking/ · /tw/democracy-needs-civic-ai/ · /tw/doom-debate/ · /tw/faq/ · /tw/gentle-bridge/ · /tw/glossary/ · /tw/inside-the-kami/ · /tw/kami/ · /tw/manifesto/ · /tw/measures/ · /tw/openclaw/ · /tw/outrage-to-overlap/ · /tw/plurality-being/ · /tw/plurality-ucsc/ · /tw/podcast/ · /tw/reimagining-ai-alignment/ · /tw/safer-sovereignty/ · /tw/self-actualisation-and-care/ · /tw/software-freedom-as-civic-care/ · /tw/sources/ · /tw/virtuous-ai/ · /tw/we-the-people-are-truly-the-superintelligence/give every Mandarin page a common-sense read at the owner's call: 344 small corrections across 39 pages and the Mandarin data (glossary definitions, comics, concept map, packs, Polis report, OpenClaw bootstrap), each checked against its English twin, where a phrase read as a calque, misled (as 「不必讀過書」 had) or said something the English does not; 中文 no longer appears anywhere (連結為中文版 → 連結為華文翻譯); no heading or ¶ id changed (anchor ledger re-recorded; parity baseline: three corrected page descriptions, plus the Mandarin narration as a required file)
34. 2026-10-01/tw/manifesto/read the Mandarin manifesto in Mandarin: /tw/manifesto/ now plays its own narration in Audrey's AI voice (ElevenLabs eleven_v4 via tts_synth.py --lang zh, which strips the publishing markup and keeps Han text, Latin terms and digits for natural Mandarin speech; the English path is byte-identical), loudness −15.99 LUFS, checked with chunked multilingual Whisper at 3.06% character error and no skipped passage
35. 2026-10-01/glossary/ · /tw/ · /tw/1/ · /tw/2/ · /tw/3/ · /tw/4/ · /tw/5/ · /tw/6/ · /tw/glossary/ · /tw/manifesto/ · /tw/measures/Mandarin wording at the owner's call: 「不必先讀過那本書」 where 「不必讀過書」 read as 'you need not be educated' (home page, Measures), 相關文獻 for 本力文獻 at the foot of each pack, 加州州長簽署 in the manifesto's later-update note, and 託管 alone no longer glosses to 託管賠償 (escrow), so 自行託管 (self-hosting) stays plain (parity baseline recaptured: the six heading ids, plus glossary-notes.json as a required file)
36. 2026-10-01/conference/sensemaking/ · /tw/conference/sensemaking/translate everything on /tw/conference/sensemaking/: the 20 Polis statements now have Taiwan Mandarin versions (keyed by comment id, beside the untouched English), carried through ranked statements, participant views, cluster standouts and history, with the remaining report prose and map labels translated; the Mandarin phone nav now reads 問答 instead of FAQ on every page
37. 2026-10-01/manifesto/ · /software-freedom-as-civic-care/ · /tw/manifesto/ · /tw/software-freedom-as-civic-care/ · /tw/we-the-people-are-truly-the-superintelligence/ · /we-the-people-are-truly-the-superintelligence/bring the ROOST note up to date in the manifesto and both essays, English and Mandarin: besides self-hostable Osprey and Coop, ROOST now runs the ROOST Model Community, which connects practitioners with the makers of open-weight safety models (OpenAI's gpt-oss-safeguard, Mistral's Shieldstral, Roblox's classifiers and others) that platforms self-host under their own policies; the note keeps its limits — shared models with local policies, not locally trained federated ones, and no outcome validation (anchor ledger re-recorded; no ¶ id moved)
38. 2026-10-01/comics/ · /glossary/ · /ja/comics/ · /openclaw/ · /tw/comics/ · /tw/glossary/ · /tw/openclaw/remove the six-dot rings at the owner's call: the Day Compass leaves the masthead (with its reading-progress script), reading-path stops go back to plain labelled links, and the home constellation keeps its slow field of lights but no longer forms the Ring of Six; chapter navigation keeps each pack's hour and Pack 4's loop back to Pack 1, and the charter now describes the site without them
39. 2026-10-01/comics/ · /glossary/ · /ja/comics/ · /openclaw/ · /tw/comics/ · /tw/glossary/ · /tw/openclaw/pair every glossary entry with its other-language term: each English entry now shows its locked Traditional Mandarin term beside it, and each Mandarin entry its English term, tagged with the right lang so it is read and hyphenated correctly; the record's locked-terms check reads the label from the paired entry
40. 2026-10-01/measures/ · /tw/measures/turn each Measures headline into an instrument card: a squared card with the pack's hue as its keel, the measure's name in the other language, its supporting diagnostics, what it refuses to reward, and its threshold and reading quoted from the page's own words, in English and Mandarin; each of the eleven named instruments now has its own anchor (#shadow-mode, #brake, …) matching its glossary id (parity baseline recaptured: the only change is those eleven ids on each page)
41. 2026-10-01/kami/ · /tw/kami/add 'Your Kami's own trace' to /kami/ and /tw/kami/: a worked example, for one fictional community pantry, of the record a room should keep from day one — keepers and successor, scope, memory files, dated checks, overrides and what changed, how to contest, and when it is reviewed or ends — each row taken from something the page already asks of the reader (parity baseline recaptured: the only change is the new heading id on each page)
42. 2026-10-01/comics/ · /glossary/ · /ja/comics/ · /openclaw/ · /tw/comics/ · /tw/glossary/ · /tw/openclaw/add the Day Compass: a six-dot ring in the interior masthead whose dots link to the six packs in their hues, the current pack's dot filled and filling with reading progress (simply filled under reduced motion or without JS); chapter navigation names each pack's hour (· Noon / ・午), Pack 4 also loops back to Pack 1 · First light, and each reading-path stop carries a small compass showing which pack it visits
43. 2026-10-01/measures/ · /tw/measures/redraw the measures map so it reads at a glance: the four-pack care cycle is a loop of cards with each handoff written on the arrow between them (a top-to-bottom sequence on phones), Pack 5 lists what it receives from the cycle, and Pack 6 frames the whole map and says in words what crosses its boundary; the sundial, circled numerals and border chips are gone, and every fact they carried is kept
44. 2026-10-01/1/ · /2/ · /3/ · /4/ · /5/ · /6/ · /tw/1/ · /tw/2/ · /tw/3/ · /tw/4/ · /tw/5/ · /tw/6/load each pack's first comic eagerly at high priority (it is the page's largest paint, and lazy loading hid it until layout), and keep the visible ¶ number inside the paragraph link's accessible name
45. 2026-10-01/sensemaker/ · /tw/sensemaker/sensemaker: give the standalone report the charter tokens, faces and colophon rules it never loaded, and scope its disclosure styles to the report so they stop restyling the Trace
46. 2026-10-01/ledger/ · /tw/ledger/QA pass across every page at ten widths: the Trace summary shows its amended date and the chip drops its duplicate amendment link; eval slots render nothing until the build fills them; bold lead-ins keep their hairline except straight under a heading; the phone masthead stacks the two names instead of pushing its tools off-screen; the margin chip never hangs past the viewport or over the comics gallery; clamped rail entries no longer leak a third line; long URLs and the agenda column wrap on 320px phones; the colophon draws its own box
47. 2026-10-01/compassion-and-ai/ · /outrage-to-overlap/ · /podcast/ · /tw/compassion-and-ai/ · /tw/outrage-to-overlap/ · /tw/podcast/footer stays readable in Lantern, and rules mark speaker turns only in transcripts: the home footer sat on --ink, which Lantern turns cream, so its pale text vanished; and every paragraph or bullet opening in bold got a rule above it, chopping prose and lists into slabs
48. 2026-10-01/sensemaker/ · /tw/sensemaker/bring the sensemaker report onto the charter: it was still loading DM Serif Display and Instrument Sans from Google on every visit, which is both a fifth and sixth voice and the only third-party request left on the site; its decorative hero glows and fading dividers are now flat
49. 2026-10-01/1/ · /5/ · /tw/1/ · /tw/5/SB 1159 as signed: California enacted it on 30 September 2026, so the four pack pages state what the law does rather than what the enrolled text would do
50. 2026-10-01/404.html · /conference/sensemaking/ · /ledger/ · /tw/conference/sensemaking/ · /tw/ledger/charter record layer: the Trace colophon with its seal, a per-page provenance ledger, machine-readable evals, an Atom feed and paragraph anchors across the manifesto, the six packs, measures and the FAQ