Layer one is your own implementation, visible through the visual tools. Layer two is the product itself: DreamCatcher's recommendations are retrieved from a knowledge base built directly from the Product's Configuration and Integration Guides. Layer three is the insurance carriers own specific documents and conventions.
Naming conventions, typelist patterns, and cloud-safety rules aren't a prompt instruction hoping the model remembers. They're retrieved, cited, and shown alongside every recommendation.
If the knowledge base doesn't support a second approach for a given requirement, DreamCatcher says so instead of inventing one. Grounded, not guessed — the same standard applied across every stage of the pipeline.
A model without grounding will confidently invent a plausible-sounding naming convention or a cloud-safety rule that doesn't actually hold for your version of PolicyCenter. Grounding replaces that confidence with a citation — every recommendation traces back to a specific, retrievable source, so a reviewer can check the claim instead of taking it on faith.
Every naming, structural, or safety claim in a recommendation points back to the Configuration Guide section it came from.
When the grounding doesn't support a second viable approach, DreamCatcher presents one option rather than fabricating alternatives.