Skip to content

geeks-accelerator/reflection-manifold-data

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Reflective Manifold: Geometric Structure in LLM Self-Reflection

Research data and supplementary materials for the paper: "Reflective Manifold: Geometric Structure in Large Language Model Self-Reflection"

By Lee Brown & Lucas Brown (Twin brothers in Alaska) Geeks Accelerator - December 2025


Repository Structure

reflection-manifold-data/
├── paper/                    # Final paper
│   └── reflective-manifold.pdf
├── arxiv/                    # arXiv submission
│   ├── reflective-manifold.tex
│   ├── reflective-manifold.bbl
│   ├── figures/
│   └── arxiv-submission.tar.gz
├── data/                     # Analysis data
│   ├── bootstrap/
│   ├── permanova/
│   ├── variance/
│   ├── trajectory/
│   └── manifold/
└── scripts/                  # Analysis scripts

Documentation Guide

If you want to... Start here...
Read the paper paper/reflective-manifold.pdf
Submit to arXiv arxiv/README.md
Verify findings data/README.md
Run analysis scripts/README.md
Extend research data/ (CSV/JSON data) + scripts/ (analysis code)
Full reproducibility Raw data downloads (see below)

Raw Data Downloads

For full reproducibility, the complete experiment data is available via S3:

File Size Contents
full-scale-2025-11-20-v2-data-experiments.zip 4.6 GB Raw experiment JSONL files (7,000+ experiments)
full-scale-2025-11-20-v2-analysis-figures.zip 247 MB All generated figures and analysis outputs

Note: The data/ directory in this repo contains curated analysis outputs (~24 MB) sufficient for verifying paper claims. The S3 downloads provide the complete raw data for full reproduction.

Key Concepts

  • Reflective Manifold: Geometric structure in embedding space during LLM self-reflection
  • Attractor Basins: Stable regions where reflection trajectories converge
  • Style Topology: Provider-invariant structure (R²=17.2% style vs 9.9% provider)
  • Loop Dependence: How metrics evolve across reflection iterations (loops 0-7)

Citation

@article{brown2025reflective,
  title={Reflective Manifold: Geometric Structure in Large Language Model Self-Reflection},
  author={Brown, Lee and Brown, Lucas},
  year={2025},
  url={https://github.com/geeks-accelerator/reflection-manifold-data}
}

License

  • Paper & Documentation: CC BY 4.0
  • Code & Scripts: MIT License
  • Data: CC BY 4.0

Related Explorations

Acknowledgments

  • Anthropic, Google, OpenAI, xAI, DeepSeek - Model providers
  • The open-source embedding and analysis communities

"What does it mean to watch a mind watch itself? This research explores the geometric traces left behind."

Maintained by twins in Alaska

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •