Refactoring 03 Config Reproducibility

Refactoring 03: Config and Reproducibility

Goal

Make experiments reproducible by centralizing configuration, controlling randomness, and recording run metadata.

Sequence

• Order: 03
• Previous: refactoring-02-dependencies-env
• Next: refactoring-04-data-io-validation

Workflow

• Centralize runtime parameters into a config object or file (YAML/TOML/argparse).
  • Success: All run parameters are set through a single config surface.
• Seed all RNGs (Python, NumPy, framework) and make the seed a first class parameter.
  • Success: Runs are repeatable with the same seed.
• Record metadata: config snapshot, git commit hash, and environment info with outputs.
  • Success: Each run output includes config and environment metadata.
• Create a consistent output directory layout for artifacts and metrics.