Skip to content
Canonical

Chunk Kind

Chunk Kind

Section titled “Chunk Kind”

Definition

Section titled “Definition”

Planned closed enumeration on content_chunks.chunk_kind capturing the structural shape of a content chunk: a heading-bounded section, a Q&A block, or a free-form paragraph. Pipeline-internal vocabulary that drives chunking + embedding + retrieval policy per chunk. CORE per §6.2 — platform-internal; clients do not extend.

Synonyms / Acronyms

Section titled “Synonyms / Acronyms”

None.

Baseline values

Section titled “Baseline values”
keylabelprovenance
heading-sectionHeading Sectioncore
qa-blockQ&A Blockcore
paragraphParagraphcore

Relationships

Section titled “Relationships”
  • content_chunks.chunk_kind — planned column (table not yet in production schema).
  • Adjacent to content_chunks.parent_chunk_id (self-FK), heading_level, heading_path[] per source §2.3.
  • Adjacent to source_documents / content_chunks (chunk-like data’s current home; content_chunks.source_document_id FK, renamed from content_item_id at ID-131 M2 — content_items itself was DROPPED at M6).
  • Layer 2: structural metadata for retrieval.
  • Layer 5: chunks are the retrieval-time unit.

Consumers

Section titled “Consumers”
  • Chunking pipeline (planned; will live in lib/extraction/ or scripts/kb_pipeline/).
  • Retrieval / RAG surfaces (MCP search + content surfaces).
  • Embedding generation policy (different chunk kinds may use different embedding strategies).
  • lib/mcp/tools/search.ts — future filter parameter when columns land.

Notes

Section titled “Notes”
  • Status: planned — column not yet in production schema.
  • Closed CV; schema migration adds new values.
  • Source §2.1 lists the initial 3 values; expansion is possible later (e.g. table-row, list-item) — review when the chunking pipeline ships.