Experiments

Experiment documentation is split by purpose:

Document	Purpose
Research Pipelines	Historical ML pipeline conventions and smoke status
Important Experiments	Consolidated benchmark and reproduction commands
Repository Layout	Active package layout and dependency rules

Recommended Workflow

1. Pick a canonical config under configs/.
2. Run genomics audit-configs --fail-on-active-legacy.
3. Run genomics audit-data --dataset-id <id> --fail-on-missing.
4. Start with a short smoke run when possible.
5. Record important results in docs/historical/IMPORTANT_EXPERIMENTS.md or a new active research note when the result should remain in the docs site.

Reproducibility Artifacts

Each integrated run should preserve:

• the original config;
• a resolved config;
• a manifest;
• metrics and model artifacts;
• enough path and dataset metadata to rerun later.