Overview
This repository collects a series of working papers on the formal structure of grounding in cognitive and artificial systems. The central thesis: hallucination is not a calibration failure but a frame misactivation— a structural diagnosis with implications for both interpretability and the design of abductive capacity in language models. Papers build on a shared architecture of NAC (Negative Abductive Capability), SSC (Selection Stability Coefficient), and the Interpretive Stack as a model of non-destructive frame accumulation.
Key Concepts
Negative Abductive Capability — the ability to withhold commitment before frame selection consolidates.
Selection Stability Coefficient — post-commitment frame coherence, defined independently of task specification T.
Frames are never deleted; suppressed frames are archived and remain accessible for re-activation.
The indexing structure over archived frames; enables structured revision without abductive regression.
Hallucination reframed as incorrect frame selection — not noise, but a deterministic structural error.
Paper Dependency Structure
Papers
Related Work
Ongoing work in extremal graph theory: lower bounds for C₄-free subgraphs of hypercubes Q₇ and Q₈, combining ILP, simulated annealing, and SAT encoding approaches.
arXiv:2603.29127 — New Lower Bounds for C₄-Free Subgraphs of Q₇ and Q₈
Presentation Slides (The Topology of Symbol Grounding)
Repository Activity (14-day window)