Core question: “What do I need to know about this source?”
A structured dossier on a single work: metadata, abstract, analysis, insights, connections to other works, bibliography, and archived source material. A compound object containing glosses, extracted principles, and typed relations to other citations. Append-only; new insights accumulate but analysis isn’t rewritten.
Citations follow the pattern Author (Year) Abbreviated Title, with an optional venue suffix , in/from Publication when venue adds important context:
[[Roy (2026) Words Without Consequence, from The Atlantic]]↑[[Chatlatanagulchai (2025) Agent READMEs]]↑The citation_slug: frontmatter field provides a short cross-referencing key (e.g., roy-2026-words-without-consequence).
A citation lead file has 7 required sections:
Frontmatter includes citation_slug: and publication_year: alongside standard fields.
Citations graduate from atomic (single file) to compound when analysis deepens:
Author (Year) Abbreviated Title/
├─ Author (Year) Abbreviated Title.md ← lead file
├─ Author (Year) Abbreviated Title — Analysis.md ← primary source analysis
├─ Author (Year) Abbreviated Title — Insights.md ← extraction candidates
├─ Author (Year) Abbreviated Title — Salience.md ← context-specific relevance bridges
├─ Renditions/
│ └─ source-title.md ← markdown copy of source
└─ Archives/
└─ citation-slug.pdf ← original binary
Sub-files use the lead file’s full name with an em-dash suffix ( — Analysis.md, ` — Insights.md, — Salience.md`). This makes all analytical depth wikilink-accessible and visible in Obsidian’s graph — garden content is synpraxis (collaborative knowledge) and must be findable. See [[Proper Obsidian Names for Garden Compound Sub-Files]]↑ for the full decision.
Sidecar files (.sidecar.md) remain invisible by design — metadata envelopes, not knowledge nodes.
| Sub-file | Role | Growth pattern |
|---|---|---|
| Analysis | Analytical breakdown: named patterns, second-order analysis, structural insights about the source | Written once per source, may be supplemented |
| Insights | Extraction candidates: each tagged with target garden form type, inference level, ghost links | Append-only as new extraction targets emerge |
| Salience | Context-specific relevance bridges: compressed arguments for why this citation matters to specific lenses, articles, or arguments. Each entry is a bridge connecting the citation to a specific audience | Append-only as the citation serves new contexts |
When both primary and secondary analysis exist, use — Analysis (Primary).md and — Analysis (Secondary).md.
is_a::[\[\[Citation Form\]\]](Citation%20Form.html)has_status::[\[\[Seed Stage\]\]](Seed%20Stage.html)in_domain::[[Domain Name]]↑cites_work_by::[[Person Name]]↑ — who wrote the cited work (subject is the citation node: “this citation cites a work by…”)cited_by::[[Other Citation]]↑ — reverse citation linkscites::[[Other Citation]]↑ — forward citation linksStructural form — captures how things relate and what we understand.
Definition from [[Deep Context as an Architecture for Captured Reasoning]], lines 68-69.
Citation Form and [[Opus Form]] share compound structure (Analysis, Insights, Salience sub-files, Renditions/, Archives/) and the em-dash suffix naming convention, but serve opposite relationships to the work:
cites_work_by::[[Person]]↑. The lead file summarizes and analyzes. The source is fixed and external. Append-only (new analysis accumulates but doesn’t rewrite).authored_by::[[Person]]↑, principal::[[Person]]↑. The lead file IS the work. The source is living and local. Revisable (the work evolves, analysis updates).Both forms share the principal-agent attribution framework (see [[Role-Specific Attribution Predicates for Opus Form]]↑) but apply it differently: Citation uses cites_work_by:: (third-person attribution); Opus uses authored_by:: and principal:: (first-person attribution).