Reflective-Ethical Engine (REE)

A minimal cognitive architecture unifying perception, action, planning, and conscious experience through trajectory selection in a fused predictive latent manifold.

Overview

REE comprises three functional components:

• E1: Deep recurrent predictor (long-horizon world model)
• E2: Fast feedforward predictor (immediate sensory grounding)
• E3: Trajectory selector (generates perception and action)

The architecture provides unified mechanistic accounts of psychiatric phenomena as failures in prediction maintenance, latent geometry, and trajectory selection.

Publication

Paper: The Reflective-Ethical Engine (REE): A Minimal Architecture for Coherent Artificial Cognition

Status: Defensive publication establishing prior art (December 2025)

License: CC-BY 4.0 International

Citation

@misc{delaharpe2025ree,
  author = {De La Harpe Golden, Daniel},
  title = {{The Reflective-Ethical Engine (REE): A Minimal Architecture 
           for Coherent Artificial Cognition}},
  year = {2025},
  publisher = {Zenodo},
  doi = {10.5281/zenodo.17859684},
  url = {https://doi.org/10.5281/zenodo.17859684}
}

Key Contributions

1. Dual-timescale predictive fusion into single temporally-displaced manifold
2. L-space as prediction-depth continuum (not discrete layers)
3. Trajectory selection as consciousness mechanism (E3 defines experience)
4. Temporal realignment in action decoding (delay compensation)
5. Unified perception-action substrate (no separate modules)
6. Computational psychiatry via manifold geometry (symptoms as distortions)
7. Minimal substrate-independent AGI architecture

Implementation Status

🚧 Coming soon: Reference implementations in JAX/Flax and PyTorch

License

This work is released under CC-BY 4.0 International.

You are free to:

• Use, implement, modify, and commercialize
• Build upon this work for any purpose
• Distribute and make available to others

Attribution required: You must give appropriate credit and indicate if changes were made.

Contact

Daniel De La Harpe Golden
Email: daniel.de.la.harpe.golden@gmail.com

Acknowledgments

This is independent scholarly work, unaffiliated with employment.

---

Note: This defensive publication establishes prior art to prevent proprietary enclosure while ensuring attribution.