This paper introduces the Rozum Framework (RF), a functional model that describes the accountable architecture of a system that can think. Accountability here is not a normative property but a technical precondition: without clear identification of the error source (external data, internal knowledge, or internal algorithms) and subsequent isolation of errors within internal structures, no learning, self-correction, or innovation is possible. The work also shows that any input processing (including thinking) is impossible without external input signals. The core finding is the reliability of output (RO) metric, a probability score quantifying output coherence and trustworthiness. RO=CI×CS×CA where CI, CS, and CA are the certainty of input, statements, and algorithms. This multiplicative structure ensures RO satisfies the mandates of the Existence axiom (survival), Accountability axiom (flaw isolation, self-recognition), and Innovation axiom (development through uncertainty). Within RF, experience is defined as the unity of internal statements and algorithms. Consciousness, which emerges as accountable self-recognition, is an executive function that verifies RO and initiates self-correction for its maximization when needed. Rozum is a conscious, substrate-independent entity that is enabled by and uses language, that is, thinks abstractly. Or within RF – consciousness that thinks.