The cognitive substrate is a research framework designed around a single premise: cognition can be expressed through repeated digital operations that follow the same structural pattern across every region and every modality. This premise is not theoretical speculation. It builds on Mountcastle's observation of the neocortex: the same computational structure, repeated everywhere.
The substrate is signal-agnostic. It receives input, routes it, generates structured marker representations, forms memory traces, and produces local predictions — all through deterministic rules, without global training, without statistical optimization, without learned embeddings. The same input always produces the same result.
"The novelty of this work lies in isolating this repeated digital operation as a standalone computational substrate, independent of global training, embeddings, or distributed parameter updates."
A neutral intake layer that converts raw input into a consistent internal representation. No semantic analysis, no feature extraction, no learned transformation. Any signal type enters through the same structural format.
A deterministic control layer that selects which downstream components process a given signal. No weights, no probabilistic scoring, no attention mechanism. Routing emerges from structural conditions, not optimization.
The core operation. A fixed array of cortical marker sets encodes the predicted state of the system across cognitive dimensions including association, pattern, prediction, relational structure, temporal properties, and epistemic stance. Markers are deterministic structural elements, not learned parameters.
Each processed signal is stored as a fully structured, interpretable memory trace containing the input, output, marker arrays, temporal key, and contextual model. Traces are retrieved by structural marker similarity, not vector distance. Salience adjusts through periodic reinforcement without global retraining.
When context is required, trace clusters are retrieved and combined into a reconstructed state model. Prediction emerges from the deterministic structural relationships present in those clusters — no generative modeling, no learned transition dynamics, no probabilistic inference.
Raw signal converted to universal internal representation.
Relevant downstream components selected. Irrelevant paths suppressed.
Cortical marker sets generated. Predicted internal state established.
Input, output, markers, temporal key, and contextual model stored.
Trace clusters retrieved. Reconstructed state model formed when required.
Updated predictive state used as reference for subsequent inputs.