This record contains the LaTeX source and compiled PDF of a research paper on the Cognitive Autonomic Framework (CAF), an architecture for distributed systems that carry and reason over an explicit model of their own semantics rather than reconstructing that structure from telemetry. The paper introduces the Runtime Semantic Tree (RST), a compiled model of a service's components, dependencies, criticality, and permitted repair actions; a gossip fabric that fuses provenance-weighted fault beliefs across peers to distinguish isolated from systemic faults; and a tiered triage and repair pipeline that admits autonomous changes only through simulate-verify-deploy verification. It states a bounded safety property for autonomous repair and an evaluation methodology covering root-cause localization, fault discrimination, runtime overhead, and behavior under RST drift. A companion reference implementation and its measurements are available in a separate record (see Related identifiers). Preliminary results reported in the paper include a doubling of correct root-cause localization when reasoning is grounded in the RST versus ungrounded prompting, and a five-fault matrix in which the verified-admission boundary escalates rather than repairs when no permitted local action exists. Author: Tobi Adeosun (Independent Researcher, Texas, USA). Status: preprint, under review.