Basic Usage

Command pattern

python -m src.cli <workflow> --config <configuration> [options]

Run commands from the repository root with the intended virtual environment active.

The execution examples below are first-run commands and write artifacts. Use a separate checkout to preserve the accepted record. Add --resume only when continuing an interrupted run with compatible artifacts; see Reproducibility.

Baseline workflow

Preview it first:

python -m src.cli baseline --config configs/base.yaml --dry-run

Then execute:

python -m src.cli baseline --config configs/base.yaml

The full baseline needs requirements.txt, not only the core CI profile, because it trains both RF and CNN baselines.

Scale-up qualification

python -m src.cli scaleup --config configs/scaleup.yaml

This selects benchmark targets, uses matched four-product coverage, builds the expanded candidate set, evaluates RF v2/CNN v2, and writes the accepted model selection. It is expensive and depends on remote mission/catalog services.

Candidate screening

python -m src.cli search --config configs/candidate_search.yaml

Before running, verify that models/production_model_selection.json identifies the intended model and that the referenced model binary is present. The command creates a frozen shortlist; it does not validate or confirm candidates.

Independent validation

python -m src.cli validate --config configs/independent_validation.yaml --stage all

This command can run many BLS searches and external queries. It consumes the frozen shortlist and the processed light curves from candidate screening.

Monitor outputs

Each workflow prints structured JSON and writes a latest run record in reports/. Inspect status, config path, counts, acceptance checks, and error messages before relying on a report.

Useful files include:

• reports/pipeline_run_latest.json;
• reports/scaleup_run_latest.json;
• reports/search_run_latest.json; and
• reports/validation_run_latest.json.

See Artifacts for the complete mapping.