The framework rests on six commitments. All other rules — AI permissions, document types, workflows — follow from these.
1. Reader authority
All philosophical meaning originates from the reader. AI may not produce, suggest, or imply interpretation without explicit request, and even then its output is marked non-authoritative.
2. Voice separation
Every claim in a note has a declared source. The framework makes it impossible to confuse primary text, language support, reader meaning, and AI organizational work.
3. Textual accountability
Claims about the text must be tied to a specific passage — book, chapter, page, or paragraph. Claims not tied to text are explicitly marked as reader-generated or AI-generated.
4. Uncertainty preservation
The framework does not resolve uncertainty the reader has not resolved. Open questions and unclear passages are tracked, not closed. "Unclear" is a legitimate epistemic state, not a failure to be corrected.
5. Asymmetric verification
Unlike mathematics, philosophical claims cannot always be locally verified. This is a permanent feature of the domain, not a deficiency. The framework accommodates it by strictly separating textual content from interpretation and requiring explicit labels on all interpretive moves.
6. AI as infrastructure, not interpreter
AI operates at the level of language access, structural observation, organization, and retrieval. It does not operate at the level of meaning, evaluation, or political classification — unless explicitly and narrowly requested.
A precise map of what AI may and may not do. When in doubt, default to the restricted side.
Permitted
Restricted
Prompt discipline
When bringing material to AI, name the mode explicitly. The mode determines what AI is permitted to do in that exchange.
Language mode
"Give me a vocabulary and grammar breakdown of this sentence. Do not comment on what it means philosophically."
Organization mode
"Organize these meaning points into the MP template. Do not rewrite any point. Add tags and suggest statuses only."
Question mode
"Generate questions that help me articulate my own understanding of this passage. Do not suggest answers or interpretations."
Resource mode
"Find me verifiable resources on [specific topic]. Give full citations. Do not use the resources to argue for a reading of the text."
Hypothesis mode — explicit only
"I am explicitly asking for an interpretive hypothesis on this passage. Give textual basis, at least one alternative reading, and mark the output [HYP]."
Six note types cover the full reading cycle. Each has a specific function and a defined boundary that prevents one type from absorbing the role of another.
1. Passage Note (PN)
Anchors a specific passage from the primary text. For English texts, includes optional inline language annotations. Entry point for all reading. Every other note type traces back to a Passage note.
Filename: [BookSlug]-PN-[ChX]-[001].md · Example: HC-PN-Pr-001.md
2. Language Table Note (LT)
A separate note containing the 4-column sentence-by-sentence table: original | breakdown & vocabulary | Chinese translation | remarks. Embedded into the Passage note via Obsidian's ![[LT-note]] syntax. For English texts this note is optional — create it only when vocabulary or grammar needs dedicated attention.
Filename: [BookSlug]-LT-[ChX]-[001].md · Example: HC-LT-Pr-001.md
3. Meaning Point Note (MP)
Stores the reader's meaning points for a passage or reading session. Each point is tagged [MP], given a status (tentative / text-supported / unclear), and linked to its source passage. AI may organize and cross-link but never rewrites a point or adds meaning content.
Filename: [BookSlug]-MP-[001].md · Example: HC-MP-001.md
4. Concept Note (CN)
Tracks a single key term or concept across the book. Does not define it once and for all. Grows as the reader encounters the term in new passages. Records evolving understanding, internal tensions, and open questions. The term's status moves from open → provisional → stable only through the reader's own decisions.
Filename: [BookSlug]-CN-[concept].md · Example: HC-CN-labor.md
5. Resource Note (RN)
Stores verified external materials: editions, translations, scholarly articles, lectures, historical background. Every entry must include citation information sufficient to verify independently. Separates primary sources, secondary scholarship, and reference material. AI may help locate; reader must verify.
Filename: [BookSlug]-RN-[slug].md · Example: HC-RN-Benhabib1996.md
6. Reading Log (LOG)
Tracks the reading process: sessions, passages read, notes created, active concepts, open questions, unresolved problems, and a running index of passage → note links. One log per book. Updated at the end of each session. The log is the reader's map of where they are and what remains open.
Filename: [BookSlug]-LOG.md · Example: HC-LOG.md
Copy these into your _templates/ folder and set them as Obsidian core templates for each note type.
Passage Note (PN)
Language Table Note (LT)
Meaning Point Note (MP)
Concept Note (CN)
Resource Note (RN)
Reading Log (LOG)
The language barrier is lower but not absent. The key discipline is to complete language support before writing meaning points — the two steps must not happen simultaneously.
[LANG] annotations directly in the Passage note.[LANG] output into the note.[MP] entries in your own words. The language annotations are available in the note, but do not re-open AI while writing meaning points.[AI-ORG].Steps 3–4 (language) and step 5 (meaning) must not happen simultaneously. Language support should reduce the access barrier; it must not shape meaning points before the reader has formed them independently. The two steps are sequential, not parallel.
For French or German texts. The language table is the primary access tool and lives in a separate LT note embedded into the Passage note. The table reduces the barrier; it does not replace contact with the original.
![[BookSlug-LT-ChX-001]] under the "Embedded language table" section.[Q] in the Remarks column. Contested terms should be preserved in the source language.A translation can carry philosophical weight invisibly. The Remarks column is where this is surfaced. When AI translates a term that has contested philosophical meaning (e.g., German Welt, Arbeit, Dasein, Öffentlichkeit), the Remarks column must flag the contest and offer alternative translations. The reader then decides whether to preserve the original term.
Meaning points are written freely first. Organization comes after. AI never writes the points — it only structures what the reader has already written.
[AI-ORG] block at the bottom records what AI did. The reader's words remain unchanged throughout the note.A meaning point can be: a conceptual observation, a noticed distinction, a question about a passage, a tentative connection between passages, a disagreement or unease, a term that seems important, or an idea that emerged while reading but is not directly from the text. All of these are welcome. None need to be polished or resolved.
Key terms are not defined once and for all. A Concept note grows with the reading and records how the reader's understanding changes over time.
[MP]. Do not attempt a final definition. Mark the status as open.[Q]. If the term seems used differently in two passages, or if a new usage challenges your working sense, record this as an open question rather than resolving it prematurely.[SCHOLAR] with full citation. Scholar readings do not replace or confirm your [MP]. They are one possible reading among others.A Concept note is never finished during active reading. Its status moves from open to provisional only when the reader feels some working stability — not when AI has supplied a definition. Stable is reserved for after the book is complete, if at all.
AI may help locate external materials, but every result must be verifiable, and resource discovery is strictly separated from philosophical judgment.
verified: yes / no / partial in the frontmatter. AI may have hallucinated a citation; never assume a resource exists until you have confirmed it.Editions and translations: which edition are you using? Are there significant differences between translations? Who translated it and when?
Scholarship: request specific names and works, not general summaries. "Who are the three most cited commentators on this book?" is a better prompt than "summarize what scholars think."
Historical context: events, debates, or intellectual traditions the text assumes. Request sourced background.
Reference entries: encyclopedia or dictionary entries on key terms, for orientation only — mark these [REF], not [SCHOLAR].
Use when you want to be prompted to think without being interpreted. AI generates questions; you answer them in your own notes, not in conversation with AI.
[Q]-tagged questions only. Good examples: "What do you think the author means by X in this sentence?" — "Does this connect to anything you read in an earlier chapter?" — "What is the source of your unease with this passage?" — "What would have to be true for this claim to hold?" — "Is your meaning point text-supported or an inference?"[MP] entries. Close the AI context before writing your answers.The risk of asking AI "what does this passage mean?" is that a plausible answer arrives, sounds authoritative, and forecloses the reader's own thinking before it has developed. Asking for questions instead keeps the interpretive space open. It also makes it easier to notice when AI has smuggled an interpretation into the phrasing of the question itself — which should be flagged and rejected.
Rules for the hardest cases: what to do when interpretation is tempting, when political labelling is requested, and when translation choices carry hidden weight.
Tracking uncertainty
Every meaning point carries a status. The reader sets it; AI may suggest but not override.
AI may suggest a status if the reader left it blank, but the reader must approve it. The reader may change a status at any time. AI may never change a status the reader has already set.
Interpretive hypotheses [HYP]
AI may only generate [HYP] when the reader explicitly requests it using those words. A valid [HYP] block must include all of the following:
A [HYP] block must never enter the reader's [MP] space. It stays in its own clearly labelled section. The reader may convert a [HYP] into an [MP] by deciding to hold it as their own tentative point — but that conversion is the reader's action, not AI's.
Political and historical classification
AI must not assign a thinker a political label (conservative, liberal, republican, totalitarian, etc.) without all four of the following conditions being met:
Even when all four conditions are met, labels must not appear in summaries, index entries, or Concept notes without explicit reader authorization.
Translation guard rules
A fresh Obsidian vault organized around books. One subfolder per book. Shared reference materials at the top level.
Naming conventions
[BookSlug] — a short fixed identifier for the book. Examples: HC for The Human Condition, OT for The Origins of Totalitarianism.
[ChX] — chapter or section code. Examples: Ch1, Pr (Prologue), Ch2S3 (Chapter 2, Section 3).
[001] — a three-digit sequential index, padded to allow sorting: 001, 002, 003…
[concept] — a lowercase slug for the concept. Examples: labor, public-realm, natality.
[slug] — a short descriptive identifier for resources. Examples: Benhabib1996, Chicago1998-edition.
Obsidian setup
Enable the Templates core plugin. Set the template folder to _templates/. Assign a hotkey to insert template.
Enable Backlinks. This lets you see which Passage notes and Concept notes link to any given MP note — useful for tracing the source of a meaning point.
Use Tags (Obsidian tags with #) for theme tags applied by AI-ORG, such as #labor or #public-realm. Do not use Obsidian tags for voice/provenance — that is handled by the inline bracket tags [TEXT], [MP], etc.
The ![[LT-note]] embed in Passage notes renders in Reading View. In Live Preview or Edit mode it appears as a linked filename.
A short illustration showing the framework in practice. Every item is labelled by voice tag. The example makes explicit what AI produced versus what the reader produced.
Passage Note — HC-PN-Pr-001
type: passage-note | book: The Human Condition | chapter: Prologue | pages: 1–6 | language: english | status: reading
"In 1957, an earth-born object made by man was launched into the universe…"
"earth-born" — compound adjective, an unusual coinage. In Arendt's German, note the distinction between Erde (earth) and Welt (world). These terms are not interchangeable in her work. The English "earth" may or may not track the German distinction. Worth checking which German term appears in the original.
"universe" — used here rather than "world." This may be a deliberate distinction. Flag as a candidate for a Concept note: earth / world / universe.
The Prologue opens with a historical event (the Sputnik launch) before moving to philosophical reflection. The shift happens at the paragraph break on p. 2. This is an observation about form — it does not claim what the shift means.
Concept notes flagged for possible creation: [[HC-CN-earth]], [[HC-CN-world]]. Language annotations generated on reader request in language mode.
Meaning Point Note — HC-MP-001
The opening sentence feels almost ironic — the launch of the satellite is presented as a triumph, but Arendt seems to be setting up a problem, not celebrating. There is something uneasy in "earth-born object made by man." Why does it matter that it is "made by man"? What else would have made it?
Is "earth-born" a philosophical category for Arendt or is it descriptive? If it is a category, what does it contrast with? Does it connect to her concept of worldliness?
Theme tags applied: #technology, #earth-world-distinction, #opening-move. Linked to [[HC-CN-world]]. No meaning points were rewritten in this session.
The reader pasted the text [TEXT] and asked AI for language help in language mode [LANG]. The reader then made a structural observation independently [STRUCT] and wrote the meaning point in their own words [MP] and posed an open question [Q]. AI organized and tagged at the end [AI-ORG]. No interpretation crossed from AI into the reader's meaning space at any point.