A Physically-Grounded, Geometrically-Mediated Architecture for Self-Originating Representation Learning: The Master Blob Framework

A design proposal for a step toward AGI: giving AI systems the ability to originate grounded representations of their own, rather than only recombining what is already latent in their pretrained weights. The paper proposes the Master Blob framework. A deformable entity in a physics simulation carries an initially untrained weight lattice that forms representation through geometric-adjacency propagation under simulated physical forces. This is connected to LLM reasoning through salience-gated memory retrieval, cross-system filtering, topological fusion of retrieved experience, and an independent proposer/judge verification loop. A three-ingredient model of conceptual innovation is used to scope what current LLM-augmentation methods do and do not address. The paper states where the origination claim stops: it applies to physically simulable domains, and general-domain origination remains open. It does not claim to be AGI or to have solved it. No component has been implemented or empirically validated. A minimal, falsifiable first prototype is specified. A companion experimental paper (Master Blob v3.1) reports a smaller, mechanistically different system and does not validate this proposal's claims. SHA-256 of the PDF: fa6b2deaf8144564cf5643ce398b5a22466202ec49674cd6121f149088521338. An OpenTimestamps proof file is included.