Reference
knowledge.chunk
domain: knowledge
A CONTENT-ADDRESSED fragment of a document: its id derives from `chunk_hash` = sha2-256(text), so identical text stored once and SHARED across document versions (git-tree). The version-relative order/index is NOT here — it lives in the knowledge.version manifest — because a chunk record is immutable + shareable. `locator` is structural (derived from content: a Stage, Section, or paragraph) so it travels with the chunk. This is the leaf the citation contract pins + re-hashes to prove no tamper.
| Field | Type | Required | Description |
|---|---|---|---|
schema | — | ✓ | |
id | string | ✓ | ds:<t>:knowledge/chunk/<hash-prefix> — derived from chunk_hash (content-addressed) |
chunk_hash | string | ✓ | sha2-256:<hex> of the chunk text — the integrity + dedup key |
doc_id | string | the document this chunk was extracted from | |
text | string | ✓ | |
locator | object | structural position: {kind: stage|section|paragraph|window, label} | |
word_count | integer | ||
tenant | string |