When Can an Agent Attribute Properties to Itself? A Typed Account of Identity, Integrity, and Freedom

An agent that reports on itself may be right, may be wrong, may lack the evidence to say - or may lack the internal structure that would make the report mean anything at all. We separate the third case from the fourth. An agent-relative property is typable when the structures its evaluation quantifies over are present in the agent's architecture; only a typable property has a verdict to be uncertain about. This is a typed integration interface applied to self-attribution; it is not a new free logic, not a new logic of de se, and not a new awareness logic. The paper's contributions are four. C1 - a typed account of self-attribution resting on three requirements (representational typability, information sufficiency, temporal certification) with an explicit dependency between them. C2 - a representational-extension result: adding a single structure can move a self-property from untypable to typed with world and evidence unchanged, a transition in the architecture alone (prop:repext). C3 - a reflexive-attribution result: the additional condition required for "r has property p" to become "I have property p" is a self-reference in the extended domain, distinct from computational reflection (cor:refl-attr). C4 - a comparative Identity / Integrity / Freedom witness family: three worked instances demonstrating that different selected self-properties can require different representational and informational support. A bounded remark connects the temporal axis to a refinement-confirmation certification layer (Bragetti, Zenodo 10.5281/zenodo.22028019) for attribution-support stabilisation under FG-persistence. --- SCOPE AND STATUS. This is a preprint. It has not been peer reviewed and is not under journal acceptance.