The Reflective Physicist: Operationalizing Universal Artificial Intelligence via System 0-1-2 Hierarchies.

This repository contains the official implementation of Horizon 2 of the Algoplexity Research Program. Its mission is to solve the Control Problem: engineering an autonomous agent capable of Zero-Shot Adaptation within the "Dancing Landscape" of complex financial markets.

The central challenge of real-world agency is the Frequency Gap—standard models possess static weights (Zero Frequency) and fast attention (High Frequency), but lack a mechanism to adapt to market regimes that change at a Medium Frequency.

To solve this, we have engineered a Nested Optimization Architecture, operationalizing Nested Learning Theory [Behrouz et al., 2025] and Coherence Theory [Williams, 2025]. The agent is not a single policy, but a hierarchy of control loops:

• Protocol Kernel: CIv2 (Autopoiesis).
• Function: Survival Initialization.
• Mechanism: Before the agent enters the market, it runs a Monte Carlo Simulation (simulate_eco_evo_transition) against synthetic "Ruin Scenarios" (Class 3 Chaos). It evolves a "Genetic Gamma" ($\gamma_0$)—a baseline paranoia level required to survive the first 100 ticks.
• Why it matters: This solves the Cold Start Problem. Unlike a standard RL agent that must die to learn, System 0 provides the Evolutionary Prior necessary for immediate viability.

• Protocol Kernel: CIv1 (Cybernetic Feedback).
• Function: Pain Avoidance.
• Mechanism: A high-frequency PID Controller that monitors the agent's own predictive error in real-time.
  • If Error > Threshold: The "Sympathetic Nervous System" activates. $\gamma$ (Uncertainty) spikes exponentially. The agent widens its wingspan to survive the shock.
  • If Error < Threshold: The "Parasympathetic Nervous System" activates. $\gamma$ relaxes linearly towards the System 2 target.
• Why it matters: This provides Kinetic Requisite Variety. It reacts to volatility faster than the neural network can think.

• Protocol Kernel: CIv5 (Structural Breaks) & CIv7 (Joint Failure).
• Function: Topological Reasoning.
• Mechanism: A Tiny Recursive Model (TRM) pre-trained on the Wolfram Computational Universe acts as the "Cognitive EEG."
  1. Sense: It maps the noisy price history to a probability distribution over Generative Rules (e.g., Rule 54 Solitons vs Rule 60 Fractals).
  2. Infer: A Reflective Gate maps this topological state to a target_gamma.
• Why it matters: It provides Strategic Direction. It tells System 1 where the baseline risk should be, allowing the agent to distinguish between "Safe Trends" (Rule 170) and "Dangerous Complexity" (Rule 54).

• main.py: The definitive Nested Learning Implementation. Contains the full QCEAAgent class, the "Fast" Homeostat, the "Slow" Physicist, and the System 0 Simulator.
• trm_expert.pth: The frozen weights of the System 2 Sensor (Horizon 1 Artifact).
• reflective_gate.pth: The learned policy mapping Topology to Uncertainty.
• notebooks/:
  • 01_H2_The_Data_Foundry.ipynb: Generating the "Nightmare Mode" theoretical benchmark.
  • 02_H2_The_Spatial_Encoder.ipynb: Training the Universal Prior for the AIT Physicist.
  • 03_H2_The_Falcon_Gauntlet.ipynb: Benchmarking the standalone System 1 (Homeostat).
  • 04_H2_The_Cybernetic_Loop.ipynb: The grand synthesis simulation.

To ensure unassailable reproducibility, this Horizon operates on an immutable scientific benchmark:

• QCEA Adaptive Agent Benchmark: The 2D "Dancing Landscape" corpus for Horizon 2.

Our work synthesizes foundational theories into a coherent engineering framework:

1. Nested Learning Theory: The mechanics of hierarchical optimization.
   • Behrouz, A., et al. (2025). Nested Learning: The Illusion of Deep Learning Architectures. NeurIPS 2025.
2. Coherence Theory: The thermodynamics of regime transitions.
   • Williams, C. F. (2025). Eco-evolutionary regime transitions as coherence loss in hereditary updating.
3. Universal Artificial Intelligence (UAI): The mathematics of optimal general intelligence.
   • Hutter, M. (2024). An Introduction to Universal Artificial Intelligence.
4. Algorithmic Information Dynamics (AID): The physics of causal structure.
   • Zenil, H. (2022). Algorithmic Information Dynamics of Cellular Automata.

If you utilize this architecture or the benchmark dataset in your research, please cite the Horizon 2 initiative:

@misc{qcea_aixi_2025,
  author = {Mak, Yeu Wen},
  title = {The Reflective Physicist: Operationalizing Universal Artificial Intelligence in Entropic Financial Environments},
  year = {2025},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/algoplexity/qcea-aixi-agent}}
}