ifp

Garden Patch Home · Glosses

Deep Context Graph Vocabulary

Deep context is a typed graph stored in plain markdown. Every document that carries predicate wikilinks is a node; every predicate wikilink is a directed, named edge. The graph emerges from the files — no database, no external index, no tooling required. An agent or human navigating the vault follows edges between nodes to build understanding.

This gloss consolidates the graph vocabulary: structural terms from the compound document decisions, new terms from compound node research, and the full semantic predicate catalog.

New Terms

Note: Any markdown document in a knowledge vault. Not source code documentation (READMEs, API docs), which has its own conventions. A note becomes a node when it participates in the typed graph.

Context node: A note that participates in the typed graph through predicate edges. Atomic or compound, garden or vault. A note with is_a:: and relates_to:: edges is a context node; a note with only YAML frontmatter and no predicates is not.

Form type: The category a node belongs to (Pattern Form, Decision Form), declared by its is_a:: edge. “Node” and “form type” are distinct: there is one Pattern Form type but many pattern nodes. When discussing individual documents, say “node” or “instance of [Form Type]”; reserve “form” for the type itself.

Typed edge: A predicate wikilink — a named, directional connection between context nodes. Written as predicate::[\[\[Target\]\]↑](../NODES.html#:~:text=Target). The predicate names the relationship type; the wikilink identifies the target. Examples: relates_to::[\[\[Deep Context Architecture\]\]](../domains/Deep%20Context%20Architecture.html), is_a::[\[\[Model Form\]\]](../forms/Model%20Form.html), derived_from::"https://example.com".

Compound node: A folder-based context node containing a lead file, optional sibling files, and optional Renditions/ and Archives/ subfolders. Generalizes “compound document” from the garden decisions to vault-wide use.

Lead file: The primary access point of a compound node — the file that gives an LLM or human the most useful context first. A single lead file serves both audiences; LLM-specific context lives in a YAML frontmatter field rather than a separate file. See [[Lead File Selection Guidance]]↑.

Terms from Compound Document Decisions

Rendition: A format-transformed markdown copy of an external source. Searchable. Lives in Renditions/. Created by conversion tools. Carries derived_from:: pointing to the canonical source. Defined by [[Renditions and Archives Replace Sources]]↑ and [[Renditions and Archives as Distinct Artifact Types]]↑.

Archive: A preserved original binary. Lives in Archives/ within a compound node folder. Named with [[[Descriptive Slugs for Archive Binaries descriptive slugs]]↑](../NODES.html#:~:text=Descriptive%20Slugs%20for%20Archive%20Binaries%7Cdescriptive%20slugs). Most don’t need sidecars. Distinct from Attachments/, which is Obsidian’s paste/drag subfolder for embedded media. Defined by [[Renditions and Archives Replace Sources]]↑ and [[Renditions and Archives as Distinct Artifact Types]]↑.

Sidecar (.sidecar.md): A metadata envelope for a binary that cannot carry its own frontmatter. Links to its binary via artifact:: (per [[Artifact Predicate for Binary Metadata]]↑) and to the canonical source via derived_from::. Only created when the binary is actively cited or needs agent discoverability. Defined by [[Sidecar Files as Metadata Envelopes]]↑.

Classification Predicates

Four predicates classify every context node in the graph:

The first two are required on every classified note. The latter two appear on garden nodes and are optional on vault types.

Semantic Predicate Catalog

Beyond classification predicates, the graph uses semantic predicates to express relationships between nodes. These are organized into four categories by the kind of relationship they describe.

Provenance Predicates (where did this come from?)

Structural Predicates (how does this relate?)

Lifecycle Predicates (what happened over time?)

Generative Predicates (what does this produce or require?)

Predicates are freeform strings. Consistency depends on ongoing vocabulary curation — awareness of what exists, periodic review for drift, consolidation of redundant terms, and clarification of ambiguous ones. The vocabulary above has stabilized through use but is not enforced by the system. See [[Predicate Vocabulary Stabilization]]↑ for the open question on when to formalize, and [[Informal Edges Poison the Graph]]↑ for the failure mode that curation prevents.

Sources

Extracted from the Graph Vocabulary gloss section of the compound nodes research note. Semantic predicate catalog from [[Deep Context as an Architecture for Captured Reasoning]], “The Predicate Grammar” section.

Relations