Inspectable cognition
Can cognitive state be represented outside hidden model weights and revisited by audit?
Research
From runtime state to cognitive ecology.
The research program began with externally observable runtime state and now studies concept formation, lifecycle regulation, open-world learning, goal-directed control, and memory consolidation.
Concepts that form and reorganize
Can attractor populations produce concepts, abstractions, and self-regulating memory depth?
Goals, questions, and consolidation
Can a runtime track what it needs, ask for evidence, and store mature structures?
Publication Map
The publication line has moved from runtime architecture into cognitive ecology.
Publications are listed by research role rather than internal numbering. Links will be added as manuscripts, supplements, and artifacts are released for public review.
State
Stateful safety and observable runtime arbitration
Early evidence that safety behavior can be treated as runtime state rather than only prompt behavior.
Architecture
Full cognitive runtime architecture
Defines the broader architecture: field state, attractors, replay, audit, and bounded dynamics.
Lifecycle
Input-supported population regulation of attractor registries
Tests attractor lifecycle, pruning, boundary conditions, and causal ablation.
Concepts
Concept emergence from attractor dynamics
Studies the relationship between attractor ecology and inspectable concept formation.
Open World
Open-world learning and cognitive ecology
Extends ADAM into wiki-scale streams, cross-version comparisons, and consolidation-ready artifacts.
Current Status
ADAM 4.0 is public-facing; ADAM 5.0 and 6.0 remain active development lines.
The public overview emphasizes ADAM 4.0 because it is the clearest current entry point for concept ecology. ADAM 5.0 and 6.0 are being developed as the next architecture layers and will be documented with bounded evidence as their validation matures.