kb-research-policy

To search a personal knowledge base, read cheap one-line index files first to build a shortlist of candidate notes, then use qmd MCP lex and vec searches with intent to balance exact and fuzzy matches before fetching full pages.

The best way to retrieve notes from a three-layer knowledge base is index-first routing: read index files to shortlist candidates across raw, wiki, and index layers, and only open full notes when the candidate list is small to avoid wasting context.

Qmd MCP lex and vec search queries use intent to balance exact and fuzzy matches across curated and external collections, allowing you to shortlist relevant wiki pages for a research query before opening full notes for validation.

Yes, you can run simple health checks and reindexing via project tasks to ensure your personal knowledge base remains discoverable and fresh for future research queries and retrieval operations.

You should avoid opening full notes during a knowledge base search until after you have shortlisted candidates using index files and qmd MCP searches, preventing premature opening of large notes and context waste.

Yes, you need to read index files first to build a shortlist before fetching full notes, as index-first routing ensures efficient retrieval and prevents opening large notes prematurely during research queries.