Notebook: Metadata usage, Metadata extraction.

Why it matters

Metadata is useful only when it is typed at ingestion and reachable at query time. This notebook does both, and shows what happens when a filter does not match the schema.

What to practise

Create a cars collection and a trucks collection with a metadata_schema of typed fields (string, integer, float, array, datetime, boolean), some marked required, and upload documents with their custom_metadata, including one that omits the required fields and should fail. Query the /generate endpoint without a filter, then with a filter_expr on content_metadata fields, a compound filter, and a filter across both collections. Turn on enable_filter_generator to derive the filter from plain language, and run the error cases: an unknown field, a wrong operator and a bad datetime.

How it fits

The typed schema is 6.4.3’s metadata enrichment, and its required fields are one of 6.4.2’s admission rules. Compare the generated filter with a manual one, and with 6.5.2’s filters that carry access decisions rather than topic.