HRPO-X (Research Prototype)

HRPO-X is a research prototype with:

• prototype utilities in hrpox/core_v2_2.py
• clean-room paper primitives (Eq3/Eq4/Eq6) in hrpox/paper_core.py
• demo-scale pipelines in hrpox/paper_pipeline.py and hrpox/paper_trainer.py

This is not production software and does not claim full paper compliance.

Roadmap: see roadmap.md.

---

Scope

Included:

• importance sampling loss with adaptive epsilon
• adaptive r_min controller
• ghost mode validation
• hash-based coordination (simulated)
• paper-aligned projection, gating, and loss primitives
• mini pipeline + trainer scaffold

Not included:

• production training system
• distributed rollout workers
• monitoring or deployment tooling
• full paper reproduction

---

Quick Start

python -m venv venv
# Windows: venv\Scripts\activate
# Unix: source venv/bin/activate
pip install -r requirements.txt
pytest tests/ -v

Run demos:

python -m hrpox
python examples/simple_demo.py

---

Structure

hrpo-x/
├── hrpox/
├── tests/
└── docs/

---

Limitations

• demo-scale pipelines only
• simulated distributed coordination
• no production infrastructure
• clean-room alignment for core equations only

---

License

MIT. See LICENSE.