Separation of roles
| Layer | Role | Boundary |
|---|---|---|
| AI capability | Produces interpretations and candidate actions | No durable authority |
| RI conditioning layer | Uses witnessed experience and relational terrain | Does not become a person or truth source |
| Human authority | Approves governed durable transitions | Cannot be inferred from model confidence |
Research surfaces
Witnessed experience
Outcomes remain attributable to their source and availability time.
Prediction error
The difference between a frozen candidate and observed outcome.
Relational terrain
Contextual dependencies and contradictions used during evaluation.
Candidate futures
Alternatives may be explored without automatic selection or action.
What must be demonstrated
RI remains a hypothesis. It must show measurable improvement against simpler baselines; adding memory alone does not earn the label.