Semantic Input
Language input enters as semantic structure before memory and control act on it.
Externally Observable Cognitive Systems
Cognitive Runtime Project
A research program for AI systems whose cognitive state is not only generated, but measured, replayed, ablated, and audited as an external runtime.
First System
ADAM is the testbed.
ADAM is an experimental cognitive runtime coupled to language models. It studies whether durable structures such as attractors, safety arbitration, field dynamics, and concepts can exist as inspectable state rather than hidden weight behavior. The model is the language substrate; the runtime is the scientific object.
Runtime Layers
Cognition as inspectable state.
Projection
Deterministic projections map semantic signals into a compact cognitive field.
Attractors
Stable structures bias future trajectories and can be inspected directly.
Field Dynamics
Runtime state evolves under input, memory, coupling, and safety pressure.
Concept Formation
Higher-order structures are studied through emergence, selectivity, and stability.
Research Arc
From runtime discipline to concept formation.
Cognitive runtime architecture
Defines the observable surface for field state, attractors, safety projection, and replay.
Determinism and bounded dynamics
Tests whether cognition and generation can remain separable, measurable, and reproducible.
Operational concept formation
Asks whether new structures can emerge, specialize, and remain stable under audit.
Evidence Standard
Evidence you can stress, replay, and remove.
Project Map
A clean entry point, with depth one click away.
cognitive-runtime.org
Observable cognition beyond hidden weights.
This site will host the project thesis, ADAM architecture notes, paper artifacts, experiment manifests, and reproducibility material for externally observable cognitive systems.