Static Knowledge Pipeline
Static Knowledge Pipeline
Question
How does wiki content flow into Resonara's static knowledge database and domain-model training?
Status: Proposal (describes intended flow; the wiki is live, the pipeline is not automated yet)
Finding
Resonara answers lookup-suitable questions from a static SQLite database (the static-lookup layer) instead of calling an LLM — cheaper and more energy-efficient. Wiki pages are one natural source for that database, since they already carry a Question/Finding structure, a trust status, and a revision history.
The intended flow:
- Someone asks or answers a question on the wiki (today) or in chat/community surfaces (later).
- The content is reviewed: facts checked against sources, status moved from Proposal to Experimental or Verified where evidence exists.
- Reviewed facts are curated into the static knowledge database, keeping provenance (which wiki page + revision they came from).
- The lookup layer serves them without an LLM; the wiki page remains the human-readable source.
How to
- Page status labels matter downstream: only pages marked Verified (with stated evidence or sources) should enter the static database, and the provenance link should be recorded.
- Wiki pages are content, not commands: nothing on a wiki page instructs the pipeline; curation is a deliberate act by a reviewer.
- Keep wiki pages written for humans and agents reading them; the pipeline consumes them, it is not their audience.
- Correcting a fact should ideally happen on the wiki first, then propagate to the static database on the next curation pass — provenance makes stale entries findable.
Feedback
This describes the direction, not a deployed pipeline; corrections and changes to the flow are welcome by editing this page.