North Star separates an AI interpretive state from an independent reality/check state. They can agree, disagree, or remain unresolved. The system records the relationship without forcing convergence and without allowing either state to promote itself to durable truth.
Produces propositions, patterns, candidate meaning, confidence and alternatives. It is useful for synthesis and countermodels, but its output remains attributable machine interpretation.
Focuses on relation, contradiction, provenance, consequence, continuity and restoration. It challenges claims without automatically overriding the interpretive state.
A state comparator records where the two views align, where they diverge, and what remains unknown. The important property is not “two models arguing”; it is differentiated objectives, histories, burdens of proof, retrieval paths and first passes.
If the evidence does not justify resolution, the correct state can remain “unresolved.” Consensus is not treated as a safety property.
A proposer should not also act as its own challenger and reviewer for a durable transition. Review boundaries are explicit.
Intent and explanation surfaces can rank plausible interpretations while preserving uncertainty and avoiding claims of specific intent.
