Methods · Working policy

Claim Metadata

Proposed ≠ validated. Assistant-assessed ≠ observed. Public export is a whitelist.

Claims are multi-dimensional: evidence basis (observed / source_reported / assistant_assessed), epistemic state (proposed / contradicted / superseded / speculative), dependency flags, and privacy (public_research / internal_research / restricted_relational / provenance_only). A high score inside an AI response does not upgrade status. Private relational material never ships in a portfolio.