Published January 10, 2026 | Version v2

OntologicalEngineering/SONAR: SONAR v1.0.0 — Initial Academic Release

Authors/Creators

Description

Published: Greene, A. (2026). Mitigating Model Collapse in Recursive 
Neurosymbolic Agents: The SONAR Benchmark for Semantic Plasticity. 
Ontological Engineering Pty Ltd. DOI: 10.5281/zenodo.18203600

Official implementation and dataset for the SONAR Benchmark protocol.

Core Implementation Details Engine: Includes the sonar_benchmark_nesy.py script, which implements the Synod-Realist-Diplomat recursive triad. Homeostasis: Provides the "Rupture" logic for detecting artifactual divergence via the τ=0.10 threshold.

Symbolic Friction: Implements the Deterministic Boolean Constraint Logic validator for actor-action grounding and propositional persistence.

Included Dataset N=30 Traces: Full execution traces for both regulated (SONAR) and ablated (Control) groups. Journal Data: Categorical logs documenting the transition from "Looping" to "Grounded" states across the 6-cycle horizon.

Legal and Intellectual Property Patent Status: Australian Provisional Patent Application No. 2026900247 was filed on 10 January 2026. The applicant has elected not to proceed with a complete application, dedicating these specific methodologies to open academic research.

License: Distributed under the Ontological Engineering Research Use License v1.0. Academic replication, regulatory audit, and independent compliance verification permitted at no charge. Commercial use requires written agreement. See LICENSE.txt.

Files

Greene_SONAR_Benchmark_2026.pdf

Files (2.2 MB)

Name Size Download all
md5:eb5650441b28bd92b453f5e90af55c49
1.5 MB Preview Download
md5:a707f14367a8a70cacdb17866b083744
716.7 kB Preview Download

Additional details

Related works

Is supplement to
Software: https://github.com/OntologicalEngineering/SONAR (URL)

Dates

Issued
2026-01-10