Integer for multi-chunk documents
MIME-like content category (e.g., "text/plain", "text/html", "application/pdf")
The Entity Document template used for vectorization
The MJ entity the vector came from (e.g., "Contacts", "Content Items")
ISO date when this vector was indexed
Composite key serialized string pointing back to the source record
Origin classification
Human-readable tags produced by the autotagging pipeline
Metadata stored alongside every vector in the shared index. Used for post-hoc filtering and provenance tracking.