Components and Integration

Long-Standing Components

These have been part of the PRIMUS/CogPrime design since the OpenCog era, now updated for Hyperon:

• PLN — uncertain reasoning backbone. In 2025, runs as quantale-annotated factor graphs for parallel, pausable, auditable belief propagation, plus backward chaining for goal-directed queries. (See PLN Full.)
• MOSES/GEO-EVO — discovers small, readable programs (controllers, macros) that slot into plans. GEO-EVO adds bidirectional guidance from current state and desired outcome along minimum-effort paths. (See MOSES Full.)
• ECAN — allocates attention (STI/LTI) over the metagraph, driving Weighted Atom Sweeps so high-value regions get priority compute. Potentially extended with incompressible-fluid-network dynamics between ECAN steps. (See ECAN Full.)
• Pattern Mining — finds reusable structures across contexts, ranked by I-surprisingness. Templates prime reasoning, suggest subgoals, and compress knowledge. (See Pattern Mining.)
• Concept Blending — creates novel ideas by merging properties from source concepts, tested by the same evaluators that rate patterns, proofs, or programs. (See Concept Blending.)

New Components (2025 Whitepaper)

These additions were introduced since the 2023 Hyperon paper. Each has a dedicated companion card under PRIMUS Advanced Components:

• MetaMo — replaces single-reward heuristics with motive geometries (families of goals/constraints) and auditable certificates. Pseudo-bimonad structure \(F = D \circ \Psi\) coupling appraisal and decision. (See MetaMo Full.)
• SubRep — CDS/PDS admission rules, Motive Decomposition Network, and weakest-sufficient decomposition via residuation. Makes subgoal discovery multi-objective, certifiable, and compositional.
• TransWeave — transfer/composition algebra ensuring cognitive steps across paradigms braid with controlled order sensitivity and carry certificates to new tasks. (See TransWeave Full.)
• WILLIAM — compression-driven pattern service on MORK. Proposes reusable motifs and acts as system-wide efficiency oracle pointing ECAN and schedulers toward highest compression/learning payoff. (See WILLIAM Full.)
• Weakness Theory — quantale-based simplicity metric native to each algorithm's algebra. Architecture-wide Occam's razor.
• Geodesic Inference — Schrödinger bridge geometry for bidirectional inference control with evidence conservation.
• Algorithmic Chemistry (ActPC-Chem) — exploratory concept: computation via self-organizing reaction networks with active predictive coding dynamics.
• QuantiMORK — proposed deep neural integration via wavelet/multiresolution DAG encoding in MORK PathMap.
• Schrödinger Bridge Learning — principled curricula from simple to accurate models via optimal transport geodesics.

How the Advances Work Together

The whitepaper (§5.13) describes a coherent flow where these components are not independent improvements but a unified enhancement:

1. Weakness-based control selects regions needing attention based on importance and uncertainty
2. WILLIAM identifies compression-worthy patterns in those regions
3. Fluid-dynamic ECAN routes attention optimally to where it's causally useful
4. PLN and ActPC-Chem refine understanding using both logical and chemical dynamics
5. Schrödinger Bridge Learning provides principled curricula from abstract to detailed models
6. MetaMo/SubRep evaluates which discovered options serve current goals
7. TransWeave determines what can be reused when facing new challenges
8. Predictive coding layers scheduled by WILLIAM and regularized by weakness theory process neural aspects

Throughout this flow, the same mathematical principles apply uniformly: weakness governs all simplicity decisions, geodesic control guides all planning, optimal transport unifies dynamics, content addressing unifies storage.