Concept ecology
L1 fragments, L2 concepts, L3 strategies, L4 abstractions, composition, decay, and reorganization.
Architecture
How ADAM is built.
ADAM is a cognitive runtime: an external system where concept objects, memory traces, goals, prediction, planning, and consolidation are represented as inspectable state. The architecture is being developed through versioned experimental lines rather than a single sealed model.
Cognitive loop
Goal state, subgoals, planning, ASK/ingest, episodic prediction, causal probes, and governor shadow.
Consolidation substrate
A long-term memory layer designed to receive mature runtime artifacts and support later retrieval.
Runtime Standard
The runtime is treated as the object of study.
ADAM separates language capability from cognitive state. The runtime supplies the measurable surface: what was active, what became stable, what decayed, what was replayed, what goal was blocked, and what evidence caused a transition.
Boundary
Language substrate is not the claim
The project studies the runtime dynamics around semantic input, not hidden weights alone.
State
Concepts have lifecycle records
Concepts and abstractions are tracked as objects with formation, activation, decay, and evidence history.
Control
Goals and governors are auditable
ADAM 5.0 adds persistent goals, subgoals, planning shadows, ASK drafts, and ecological diagnostics.
Memory
Consolidation is a separate substrate
ADAM 6.0 is being developed as a bridge from runtime ecology to longer-term structured memory.
Version Boundary
Different versions answer different questions.
The version line is intentional. ADAM 4.0 asks whether concept ecology forms. ADAM 5.0 asks whether that ecology can be connected to goals, planning, questions, and governance. ADAM 6.0 asks whether mature runtime artifacts can be consolidated without erasing their provenance.