Triple Separation
Architectural rule separating raw text, normalized claims, and explicit relations
Triple separation is the architectural rule in a Dialectical Graph that keeps three layers distinct: original text is preserved as quotable evidence spans; claims are decomposed into normalized propositions; and relations are stored as explicit argumentative bindings.
This separation prevents a common failure mode in RAG: excerpt truncation. When text is retrieved and summarized without preserving the original wording, attribution can silently disappear. Triple separation ensures that you can always trace a claim back to the exact words and the source that produced them.
It also enables knowledge synthesis by making structure queryable. You can ask "what attacks this claim?" or "what definitions does this rely on?" because those relations are stored, not implied.