Extropy as a Metric of Cognitive Novelty: Toward a Computational Framework for Signal Recognition and Extropic Semantic Codecs
Content filters for language models are usually binary: an output either matches a disallowed pattern or it does not. This preprint proposes a graded alternative, extropy, defined as the product of novelty (distance from an expected manifold), contextual coherence and logical integrity, so that useful novelty can be surfaced rather than suppressed. Chapters 1–3 specify the textual path: a tangent-residual novelty score and its cheap angular proxy (the Local Extropy Index), a Jensen–Shannon transformational delta, an exponentially smoothed trust vector and several anti-noise measures. Chapter 4 operationalizes the same product as an Extropic Semantic Codec on a Gray–Scott reaction–diffusion field. A packet of a few hundred bytes carries an ontology pointer, a manifold anchor, an extropic residual and a superlevel H0 persistence barcode, and the receiver replays the field from a shared prior. On the MORPH-OS reference suite, L1 packets average 472 bytes against about 11.3 kB for a PNG at 128×128 and preserve working-threshold β0 in all six homology test cases. Every number in the measured tables traces to a named file in the public MIT-licensed repository (github.com/IAmM3ta/MORPH-OS), and the homology tables regenerate exactly from esc/run_homology.py. Chapter 5 states an adversarial-hardening policy as a design without measurements. Chapter 6 gives thought experiments and next measurements, including why the codec is not an encryption scheme.
Authors
- Robert F. Thomas (ORCID: https://orcid.org/0009-0006-2901-6294)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-21
- DOI
- https://doi.org/10.5281/zenodo.24417495
- Primary Topic
- Topological and Geometric Data Analysis
- Type
- preprint