Skip to content

Research 0703: AI Materializer CLI Helper Sweep

Question

After the import-bootstrap sweep, which active single-table AI scripts should use the shared CLI helper layer from ADR-0680?

Inputs Reviewed

  • ai/scripts/materialize_saliency_features.py
  • ai/scripts/materialize_second_opinion_features.py
  • ai/scripts/materialize_mos_labels.py
  • ai/scripts/signal_mix_audit.py
  • ai/scripts/feature_correlation.py
  • ai/src/aiutils/cli_helpers.py
  • ai/tests/test_materialize_saliency_features.py
  • ai/tests/test_second_opinion_features.py
  • ai/tests/test_materialize_mos_labels.py
  • ai/tests/test_signal_mix_audit.py
  • ai/tests/test_feature_correlation.py

Findings

These scripts still built argparse.ArgumentParser instances directly and captured raw sys.argv locally even though the batch wrappers now use make_argument_parser() and collect_cli_argv(). That leaves the same materializer stack with two parser/provenance idioms.

The shared helper is intentionally small: it does not change CLI options, manifest schemas, materializer behavior, or report payloads. It only standardizes parser formatting and raw-argv capture for ADR-0661 provenance.

Scope Chosen

Migrate the three materializers plus signal-mix and feature-correlation reports to make_argument_parser() and collect_cli_argv().

Reproducer / Smoke

.venv/bin/python -m pytest \
  ai/tests/test_materialize_saliency_features.py \
  ai/tests/test_second_opinion_features.py \
  ai/tests/test_materialize_mos_labels.py \
  ai/tests/test_signal_mix_audit.py \
  ai/tests/test_feature_correlation.py -q

Limits

This does not migrate every historical AI training/export script. Those can move to the helper when their owning model or report surface is already being touched.