Module schema 

Arrow schemas for the graph store.

Three foundational tables:

• Triples: subject/predicate/object quads with provenance
• Embeddings: entity vectors (FixedSizeList<f32>)
• Metadata: per-entity access tracking

The triples schema includes an optional layer column (UInt8) for sub-partitioning within namespaces. When layers are not needed, the column is still present but set to 0.

Modules

chunk_col

Named column indices for the Chunks schema (fine-grained provenance).

col

Named column indices for the Triples schema. Use these instead of hardcoded integers when accessing RecordBatch columns.

Constants

CHUNKS_SCHEMA_VERSION

Current schema version for the Chunks table.

DEFAULT_EMBEDDING_DIM

Default embedding dimension (all-MiniLM-L6-v2 = 384, but 768 is future-proof).

TRIPLES_SCHEMA_VERSION

Current schema version for the Triples table.

Functions

chunks_schema

Schema for the Chunks table — fine-grained document provenance.

embeddings_schema

Schema for the Embeddings table — vector representations of entities.

embeddings_schema_with_dim

Embeddings schema with a custom vector dimension.

metadata_schema

Schema for the Metadata table — per-entity access tracking.

normalize_to_current

Normalize a RecordBatch from an older schema version to the current version.

triples_schema

Schema for the Triples table — the core knowledge representation.