airtable-knowledge-extract

Airtable knowledge extraction transforms Airtable bases into vault knowledge by mapping records to linked entity pages with wikilinks. It performs per-base cluster extraction and full-cluster subgraph traversal to seed the vault with structured historical knowledge.

Airtable knowledge extraction uses a Sonnet worker with a per-batch reviewer at a 95% quality gate. It enforces mandatory human-in-the-loop re-anchoring on every new base and runs a deterministic review cycle to gate import quality before writing to the cache.

Airtable extraction operates across bases using a per-base vertical-slice approach with a per-run cache. However, it enforces one-base-at-a-time reviews with mandatory human-in-the-loop re-anchoring on every new base to maintain knowledge graph coherence.

Airtable knowledge extraction stages outputs in the cache directory and appends an outcome log for audit. This deterministic review cycle produces reproducible results, allowing users to verify imported entity pages and track extraction outcomes per base.

Airtable knowledge extraction includes legacy-link detection and per-base cleanup to keep the knowledge graph coherent. It identifies existing connections across bases and resolves them during the subgraph traversal process to maintain link integrity.