Abstraction engine
A background worker that clusters experience events into progressively higher-order concepts using embedding-based similarity. Produces a five-level compression ladder: experience - pattern - concept - principle - worldview. Vendor-agnostic at every level.
The abstraction engine runs as a background worker separate from the real-time cognitive loop, consuming consolidated experience events and grouping them by embedding similarity into progressively higher-order representations. The five-level ladder it produces - experience, pattern, concept, principle, worldview - reflects increasing generality and compression. A single experience event records what happened; a pattern captures that a class of events recurs; a concept names the shared structure; a principle states the generalizable rule; a worldview represents the broadest belief about how a domain operates. Each level feeds the level above it. The engine is vendor-agnostic by design: the embedding models, clustering algorithms, and storage backends can be swapped without changing the ladder structure or the downstream systems that consume abstractions. This portability matters because the abstraction quality is only as good as the embedding model, and models improve over time.