Method and trust
Why Imaginal AI is an advisory infrastructure
Missions, role-defined councils, separate corpora, commitments, and longitudinal context—not a personality chatbot. Imaginal AI is built around explicit corpus boundaries, role-defined disagreement, and accountable action rather than a single all-purpose personality.
The architecture
Four boundaries make the product inspectable.
Mentor boundary
Every mentor has a stable ID, registered works, an isolated search engine, and a release-map entry. Missing entries fail closed.
Evidence boundary
The answer prompt receives only retrieved passages for the selected mentor. Citations must belong to that returned set.
Council boundary
Members answer independently before synthesis. The Meta-Advisor may compare positions but may not invent consensus or citations.
Action boundary
Counsel becomes useful only when the user chooses a commitment, owner, date, and review signal. The system never claims the outcome.
Rights and provenance
A source must be governable before it becomes retrieval infrastructure.
Each registered edition records its provider, work, territorial review, and release status. Five mentors remain blocked because the available edition or translation lacks the required production evidence. Their profiles stay visible for transparency, but they are noindexed and unavailable to production retrieval.
Private custom mentors follow a different boundary: the user attests to permission, files remain user-scoped, indexing occurs in a private resource, and deletion removes the associated objects and metadata.
Product boundary
The system is educational decision support.
It is not the historical person, a substitute for professional advice, or a mechanism for guaranteeing business, psychological, financial, medical, legal, or personal outcomes. Its job is to improve the quality and traceability of the user's deliberation.