Choosing the right corpus for Kedapix is a product decision, not a data chore. We now ship a trustedSources toggle inside every topic so you can gate which URLs feed the LLM router.
The rubric#
Signal density#
We look for assets where 80%+ of the paragraphs translate directly into decisions. That includes board decks, working memos, and PR FAQ docs with concrete counter-metrics.
Expert adjacency#
If a document references names, job titles, and dates that our agents can verify via Supabase, we treat it as first-party. Everything else goes through a second pass with higher thresholds.
Refresh cadence#
Each tag in Kedapix listTags() maps to review cadences:
product-updates: weekly because UI screenshots drift quickly.ai-research: monthly, aligned with arXiv drops.infrastructure: quarterly, tied to cost reviews.
We will keep sharing the scoring sheets so you can remix them for your own Kedapix topics.
