Research 0702: AI Materializer Bootstrap Sweep¶
Question¶
Which scripts should migrate immediately after the new AI bootstrap helper lands so the batch runners and their single-table tools do not use different import patterns?
Inputs Reviewed¶
ai/scripts/materialize_saliency_features.pyai/scripts/materialize_second_opinion_features.pyai/scripts/materialize_mos_labels.pyai/scripts/signal_mix_audit.pyai/scripts/feature_correlation.pyai/scripts/_script_bootstrap.pyai/tests/test_materialize_saliency_features.pyai/tests/test_second_opinion_features.pyai/tests/test_materialize_mos_labels.pyai/tests/test_signal_mix_audit.pyai/tests/test_feature_correlation.py
Findings¶
The batch manifest runners now use bootstrap_ai_script(), but the underlying single-table materializers still had local Path(__file__) plus sys.path.insert(...) blocks. That split would make future edits harder to review because wrapper scripts and their worker scripts would bootstrap imports differently even though they live in the same materialization stack.
signal_mix_audit.py and feature_correlation.py are in the same active AI artifact-analysis path: both emit durable reports with run provenance and both use the same repo-root plus ai/src import shape. Migrating them in the same sweep keeps the signal-audit tools aligned with the materializer stack without touching model training behavior.
Scope Chosen¶
Migrate these five scripts to bootstrap_ai_script(__file__) only. Keep their schemas, report payloads, materializer behavior, and CLI arguments unchanged.
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¶
The remaining historical AI scripts can migrate later when their owning workstream is already touched. This sweep deliberately avoids training/export entrypoints with larger dependency surfaces.