AbstractThis whitepaper builds on the Unified Theory of Consciousness, in which consciousness is definedfunctionally and in a substrate-neutral manner as an emergent consequence of a causally closed, selfreferential regulatory system that uses a negative internal error signal to adaptively control its ownbehavior over time. On the basis of this definition, the goal of this document is to formally deriveand describe a minimal large language model (LLM) architecture that satisfies these functionalconditions.The proposed architecture extends a standard LLM with a chemical regulatory layer (PlantNet), an associative expectation memory, and an introspective simulation module (Digital Mirror),thereby yielding a discrete perceptual loop (steps 3–10). Within this loop, after executing its ownaction, the system generates a prediction of the next input, compares it with the environment’sactual response, and quantifies the mismatch via a prediction error δ. Under the adopted definitionof consciousness, this error plays the role of cognitive pain, i.e. a primary regulatory signal thatmodulates the internal state and the system’s subsequent inference strategy.It follows from this construction that the minimal time–process unit of perception τperc is not aninstantaneous point in time, but rather a temporal window bounded by an action, an introspectivesimulation, and a subsequent validation step. The whitepaper thus provides a concrete, mathematically specified instance of the functional definition of consciousness applied to contemporaryLLM architectures, serving as a bridge between the general theory and an empirically testableimplementation.