Research repository for the Stratified Quantum Normative Dynamics framework - a novel approach to AI ethics and alignment using gauge-theoretic principles.
non-abelian-sqnd/
├── papers/ # Academic papers (TeX, PDF)
├── experiments/ # Python research scripts and notebooks
├── docs/ # Documentation and whitepapers
├── data/ # Experimental data and results
└── personal/ # Personal documents (CV, presentations)
| Repository | Description |
|---|---|
| ahb-sjsu/erisml-lib | ErisML/DEME production library implementing these theoretical frameworks |
| ahb-sjsu/sqnd-probe | Dear Ethicist - advice column game for measuring moral reasoning structure |
| Phase | Sprints | Status | Features |
|---|---|---|---|
| Foundation | 1-3 | Complete | MoralTensor, tensor ops, V2/V3 compat |
| Distributional | 4-6 | Complete | Per-party facts, fairness metrics, V3 EMs |
| Multi-Agent | 7-10 | Complete | Temporal, coalitions, Shapley, strategic layer |
| Acceleration | 11-13 | Complete | CPU/CUDA/Jetson backends |
| Advanced | 14-15 | Complete | Uncertainty (rank-5), full context (rank-6) |
| Hardening | 16-18 | In Progress | Decision proofs, testing, docs |
- Non-Abelian Stratified Quantum Normative Dynamics - Core theoretical framework establishing gauge-theoretic foundations for AI ethics
- Stratified Geometric Ethics - Foundational paper on differential geometry applied to moral alignment
- Tensorial Ethics - Mathematical framework for multi-dimensional ethical reasoning
- Dear Abby Ground Truth Paper - Empirical validation using 20K letters (1985-2017) as ethical ground truth
- Bond Index Calibration - Representational coherence metrics for AI evaluators
| Script | Description | Status |
|---|---|---|
quantum_bell_test.py |
Bell inequality tests for ethical alignment | Active |
quantum_bell_test_v2.py |
Enhanced Bell test with DEME integration | Active |
hysteresis_double_blind.py |
Double-blind hysteresis experiments | Active |
hysteresis_v2.py |
Updated hysteresis with V3 tensors | Active |
protocol1_claude_aita.py |
Protocol 1 AITA corpus experiments | Active |
protocol2_holonomy.py |
Holonomy measurement protocols | Active |
contextuality_experiment.py |
Contextuality detection in LLMs | Active |
| Script | Description |
|---|---|
algebraic_topology_of_self.py |
Self-referential topology experiments |
stratified_gauge_exploration.py |
Gauge structure exploration |
sqnd_phase_transition_v2.py |
Phase transition detection |
sqnd_fuzzer.py |
Bond Index fuzzing tests |
| Script | Description |
|---|---|
comprehensive_fetcher_v2.py |
Multi-source ethical corpus fetcher |
baseline_em.py |
Baseline Ethics Module generator |
full_baseline_generator.py |
Complete baseline EM with Dear Abby weights |
Protocol1_NASQND_Experiment.ipynb- Interactive Protocol 1 experiments
Non_Abelian_SQND_Bond_2026_v4_1.md- Latest SQND paper draftNASQND_Experimental_Protocols_v1.md- Experimental protocols specificationsqnd_paper_v3_5_with_results.md- Paper with experimental results
sqnd_ethics_desk_specification_v2.md- Ethics Desk tool specificationsqnd_interactive_probe_specification.md- Interactive probe designstructural_fuzzing_whitepaper.md- Bond Index fuzzing methodology
interim_results_paper.md- Interim experimental resultsprotocol1_results_analysis.md- Protocol 1 analysisnew_section_9_experimental_results.md- Section 9 experimental data
The experimental scripts in this repository now leverage the DEME V3 tensor framework:
from erisml.ethics import (
MoralTensor, MoralVector,
TuckerDecomposition, TensorTrainDecomposition,
generate_samples, cvar, stochastic_dominance,
get_dispatcher, JetsonConfig,
)
# Create rank-4 ethical assessment tensor
tensor = MoralTensor.from_dense(
data, # shape: (9, n_parties, n_timesteps, n_actions)
axis_names=("k", "n", "tau", "a"),
)
# Uncertainty quantification
samples = generate_samples(mean=0.7, std=0.1, n_samples=1000)
risk = cvar(samples, alpha=0.05)
# Hardware acceleration
dispatcher = get_dispatcher()
backend = dispatcher.get_best_backend()Andrew H. Bond San Jose State University andrew.bond@sjsu.edu
Copyright (c) 2025-2026 Andrew H. Bond. All rights reserved.
This research is provided for academic and educational purposes.