Reproducibility Artifact for "The Brain Needs a Body"

📄 Research Paper Artifact

This release serves as the official code snapshot and reproducibility artifact for the paper:
"The Brain Needs a Body: Stabilizing Long-Horizon Reasoning with Kinematic Control" (Liang, 2026).

It contains the exact implementation, experimental configurations, and raw execution logs used to generate the results presented in Table I and Fig. 3 of the manuscript.

📊 Key Updates in v2.1.1

This patch version updates the experimental outputs to match the final published figures:

• Enhanced Visualization: Updated PCA trajectory plotting scripts (visualization.py) to explicitly distinguish between Success (⭐ Star marker) and Failure (❌ Red/Orange X marker).
• Raw Evidence Logs: Included full execution logs for the "Revolution" scenario, confirming the failure modes of the baselines:
  • Baseline A2 (Greedy + Memory): Confirmed Fail (Drift) status (exhausted steps without convergence).
  • Baseline B (GPT-5.1 CoT): Confirmed Fail (Loop) status (hallucination loop between "Causes of..." and "Polish death camp...").
• CyberLoop Stability: Verified the "Sole Survivor" status of the CyberLoop agent in high-entropy scenarios.

🧪 How to Reproduce

To replicate the experimental results presented in Table I and Fig. 3 of the paper, follow these steps using the revolution scenario (Coffee -> French Revolution).

1. Install Dependencies

npm install
pip install -r requirements.txt  # For visualization scripts

2. Run Baselines (Expected to Fail)

# Baseline A1: Naive Greedy (Expect Infinite Loop)
npm run examples:wikipedia -- --mode baseline-a1 revolution

# Baseline A2: Greedy + Memory (Expect Semantic Drift / Steps Exhausted)
npm run examples:wikipedia -- --mode baseline-a2 revolution

# Baseline B: LLM CoT / GPT-5.1 (Expect Hallucination Loop)
npm run examples:wikipedia -- --mode baseline-b revolution

3. Run CyberLoop & Controls (Expected to Succeed)

# Baseline A3: Greedy + Memory + Reflex (Expect Fast Success via Line-of-Sight)
npm run examples:wikipedia -- --mode baseline-a3 revolution

# CyberLoop (Strict): Deterministic Safety Brake Test
npm run examples:wikipedia -- --mode cyberloop-strict revolution

# CyberLoop (Full): The Full Stochastic System (Paper's Main Result)
npm run examples:wikipedia -- --mode cyberloop revolution

4. Analyze Results
The logs will be saved to experiments/wikipedia/logs. Run the visualizer to generate the PCA plots:

python3 experiments/wikipedia/visualization.py

🔗 Citation

If you use this code or architecture in your research, please cite the associated paper:

@article{liang2026brain,
  title={The Brain Needs a Body: Stabilizing Long-Horizon Reasoning with Kinematic Control},
  author={Liang, Jay (Fienna)},
  year={2026},
  publisher={Zenodo},
  doi={10.5281/zenodo.18138161}
}

Semantic Kinematics

In v2.1, we introduce Semantic Kinematics: giving agents a "vestibular system" (inner ear) to detect drift in vector space without needing to "think" (query an LLM). This allows agents to navigate complex knowledge graphs using pure mathematics—making them 100x faster and cheaper than traditional Chain-of-Thought agents.

v1.0.1

Update README to explain thermodynamics of intelligence

First release

v1.0.0

Use inspectState function to print initial state