RITAM (Research Infrastructure for Trustworthy AI Mechanisms) is a governed cognition substrate — a runtime layer that sits beneath an AI application and holds its cognitive state under explicit governance. Current AI systems silently accumulate contradictions. A system told conflicting facts has no principled response: it overwrites, ignores, or crashes. RITAM addresses this at the substrate level, before any application logic runs. The substrate implements nine primitives, each justified by a concrete failure mode that appears when the primitive is absent: State, Memory, Ontology, Governance, Epistemic, Coordination, Temporal, Observation, and Repair. The founding principle is that governance must precede persistence — beliefs are governed at admission, not cleaned up afterward. v1.1.1 includes 146 tests across adversarial, integration, and buildability scenarios. Key findings: governance changes outcomes (a governed substrate produces measurably different results from an ungoverned baseline); all nine primitives are load-bearing (removing any one causes observable failure); and the specification is transferable (five independent AI systems reproduced the runtime from the specification alone, 60/60 tests passing). Three consumer types are demonstrated: governed research assistant, governed decision log, and governed agent memory — confirming the substrate is domain-agnostic. This is a research prototype. The architecture is the contribution; the implementation demonstrates that the architecture is buildable, testable, and reproducible by independent builders.