Conceptual Engine and Language Models
Reference: This page is a source-based technical synthesis of the LaTeX chapter cited below. For exact syntax, availability or ABI signatures, verify the versioned source, manifest and executable tests.
Scope and source boundary
An LLM may rank or propose conceptual candidates; its linguistic output is not itself logical proof.
Tensor and vector representations are derived numerical projections of semantic or runtime state. The source explains how embeddings, latent retrieval and resident buffers accelerate candidate selection, while logical decisions continue to rely on semantic authority. Numerical scoring and H-Logic proof therefore have distinct validity and lifecycle rules.
Engineering rules
- Record which concept or fact each vector represents.
- Invalidate or refresh numerical projections after source model changes.
- Use similarity to prioritize evaluation, not to invent a semantic relation.
- Manage tensor provider, memory and device lifetime explicitly.
Chapter outline (original LaTeX headings)
- Candidate Contract
- Composite Scoring
- Context Integration
- Example
LaTeX provenance
Primary chapter: logicells-neuro-symbolic-architecture-guide/chapters/neural-intuition-contract.tex.