ONN project status report -- codebase, tests, and paper readiness
961 words4 min read
Project overview
ONN-ES (Ontology Neural Networks with Evolution Strategy) is an edge-centric graph neural network framework combining topological constraints with evolution strategy (CMA-ES) for inference.
The core architecture follows a Trinity of Operators layout:
SEGO (Semantic Graph Observer) -- perception and feature extraction
LOGOS (Logical Ontology Graph Optimization Solver) -- correction and stabilisation
IMAGO (Intent Modeling via Graph Optimization) -- planning and path generation
ORTSF (Ontology Real-Time Stability Fabric) -- delay-compensated control
Message changed ('All checks passed' vs 'Stable state')
Low
Analysis. Four failures trace to LOGOS loss normalisation switching from sum to mean without updating the test expectations. Three are simple string-casing mismatches. No core-logic bugs.
Paper readiness
Experiment data
Experiment
CSV
Rows
Status
B1 -- Loss Dynamics
B1_loss_dynamics_steps.csv
2,664
Done
B3 -- Seed Robustness
B3_seed_robustness.csv
41 (20 seeds x 2 versions)
Done
B4 -- Scaling Study
B4_scaling_study.csv
31 (N = 2..32)
Done
B6 -- Ablation Study
B6_ablation_study.csv
51 (6 configs x 5 seeds)
Done
Figures (11 / 12)
Figure
Status
A1 -- System Overview
Done
A2 -- Node Anatomy
Done
A3 -- LOGOS Loop
Done
A4 -- Tradeoff Triangle
Done
A5 -- Fix Map
Done
A6 -- Meta-LOGOS Gate
Done
B1 -- Loss Dynamics
Done
B2 -- Gradient Stability
Done
B3 -- Seed Robustness
Done
B4 -- Scaling Study
Done
B5 -- (TBD)
Missing
B6 -- Ablation Study
Done
LaTeX manuscript
Item
Status
IEEE template (IEEEtran.cls)
Ready
onn_main.tex
Ready
01_introduction.tex
Ready
02_related_work.tex
Ready
03_methodology.tex
Ready
04_experiments.tex
Ready
05_discussion.tex
Ready
06_conclusion.tex
Unconfirmed
references.bib
Ready
Reproduction scripts
All four experiment scripts (paper_experiment_B1/B3/B4/B6.py), the combined plotter (paper_plot_B1_to_B6.py), the logger utility (paper_collect.py), and the orchestration script (paper_run_experiments.py) are complete.