Research 0704: AI Corpus Adapter CLI Helper Sweep¶
Question¶
Which corpus JSONL adapters still duplicate direct-invocation path setup, parser construction, and raw-argument provenance capture after ADR-0680 and ADR-0681 introduced shared helpers?
Inputs Reviewed¶
ai/scripts/chug_to_corpus_jsonl.pyai/scripts/konvid_1k_to_corpus_jsonl.pyai/scripts/konvid_150k_to_corpus_jsonl.pyai/scripts/youtube_ugc_to_corpus_jsonl.pyai/scripts/waterloo_ivc_to_corpus_jsonl.pyai/scripts/lsvq_to_corpus_jsonl.pyai/scripts/live_vqc_to_corpus_jsonl.pyai/scripts/bvi_dvc_to_corpus_jsonl.pyai/src/aiutils/cli_helpers.pyai/scripts/_script_bootstrap.py
Findings¶
The MOS/feature materializer family already moved to the shared helper shape, but the corpus JSONL adapters still carried older idioms:
- direct
argparse.ArgumentParser(...)creation in every adapter; - local
sys.argv[1:]capture for manifest provenance; - repeated
Path(__file__).resolve().parents[2]repo-root derivation; - in the BVI-DVC adapter, hand-written
sys.path.insert(...)blocks fortools/vmaf-tune/srcandai/src.
The helpers are behavior-preserving for these scripts. They keep the same CLI flags, output schemas, default paths, and manifest schemas while making direct script execution and run-provenance capture match the rest of the AI stack.
Scope Chosen¶
Migrate the active corpus JSONL adapter group to:
bootstrap_ai_script(__file__)orinclude_vmaf_tune_src=Truewhere the adapter importsvmaftune;make_argument_parser(...);collect_cli_argv(argv);- bootstrap-provided
SCRIPT_PATH/REPO_ROOTvalues for manifest provenance.
Reproducer / Smoke¶
.venv/bin/python -m pytest \
ai/tests/test_chug.py \
ai/tests/test_konvid_1k.py \
ai/tests/test_konvid_150k.py \
ai/tests/test_youtube_ugc.py \
ai/tests/test_waterloo_ivc.py \
ai/tests/test_lsvq.py \
ai/tests/test_live_vqc.py \
ai/tests/test_legacy_corpus_extraction_manifests.py -q
Limits¶
This sweep does not migrate every training/export/evaluation script. The remaining scripts should move in owner-family batches so tests and report schemas stay easy to reason about.