​
Knowledge Base

Snowglobe treats your knowledge base as the authoritative corpus for what the chatbot should know. The simulation engine mines topics from these files to build scenarios and evaluate how well your agent stays on-policy.

​
What to Include

  • Every external document your production model references, especially retrieval-augmented generation (RAG) sources.
  • Policy, FAQ, and troubleshooting guides that provide ground truth for expected answers.
  • Updated copies of any material your team regularly refreshes in production.
Keeping the corpus incomplete or stale limits the variety of simulations Snowglobe can generate.

​
Supported Formats

Plain-text formats only: .md, .txt, .json, or .csv. Convert PDFs, slides, or rich docs into one of these formats before uploading so the engine can parse the content.

​
Why It Matters

  • Scenario coverage: Snowglobe mines topics directly from the knowledge base to craft simulation prompts.
  • Hallucination scoring: The hallucination metric checks responses against these files as ground truth, so missing docs can mask incorrect answers.
  • Production parity: Matching the files your agent actually retrieves keeps simulation outcomes aligned with real conversations.
Refresh the knowledge base whenever your production content changes to keep simulations and metrics trustworthy.