absorb

To build a knowledge base from scattered markdown and text documents, ingest a target directory to extract facts, decisions, and timelines. This process structures unstructured content into an updated knowledge base with provenance tracking for every reflected item.

The best way to detect stale information and conflicts during knowledge extraction is cross-validating new content against existing sink documents. This approach preserves conflicts for user judgment instead of auto-resolving them and identifies outdated data.

To perform bulk document ingestion without reprocessing files, use an incremental run via an absorb-log. This idempotent method skips already-absorbed markdown and text files unless explicitly forced to reprocess the entire directory.

No, knowledge base maintenance does not automatically resolve conflicting project decisions. It detects duplicate or conflicting information during cross-validation and preserves these conflicts explicitly for user judgment rather than auto-resolving.

Useful knowledge when converting project documents into a structured format must pass a So What filtering quality gate. This keeps only facts, rules, and insights likely to be useful to a future teammate after 6 months.