Contribute

PAUL Open is an open research project. We welcome contributions to our datasets, evaluation suites, and training infrastructure.

Development Setup

We use uv for Python package management and environment isolation. Ensure you have uv installed before beginning.

$ git clone https://github.com/foundrypaul-cloud/paul-open.git$ cd paul-open$ uv sync --extra all$ uv run pytest

Code Standards

All contributions must pass our strict CI pipeline. Run the following commands locally before submitting a Pull Request:

# Format code
uv run ruff format .

# Run linter
uv run ruff check .

# Static type checking
uv run mypy src/

# Run unit tests
uv run pytest

Dataset Contributions

All dataset contributions must pass the Tri-Tier Validation framework:

• Tier 1: Must pass schema validation (JSON/JSONL).
• Tier 2: Must pass heuristic screening (no verbosity bias, correct structure).
• Tier 3: Must undergo human-expert review (recorded in metadata).
• Anti-Leakage: Must pass scripts/check_leakage.py against the frozen evaluation suites.