build-manifest

name	build-manifest
description	Build a dataset manifest and validate it against the manifest schema.
argument-hint	<dataset-name> [--root <data-root>] [--out <manifest-path>]

You are about to build a dataset manifest. The manifest is the unified internal representation that every adapter writes (implementation plan task 2.3).

Steps:

1. Read data/README.md and docs/datasets/dataset_license_matrix.md to confirm the dataset is allowed and how to access it.
2. Run:

   python scripts/build_manifest.py --dataset $1 \
       --root ${RP_DETR_DATA_ROOT:-data/raw}/$1 \
       --out data/manifests/$1.jsonl
   

   If $2 / $3 are passed, override --root / --out accordingly.
3. Validate:

   python -m rp_detr.cli.validate_predictions \
       --schema schemas/dataset_manifest.schema.json \
       data/manifests/$1.jsonl || true
   python -m rp_detr.cli.manifest_stats data/manifests/$1.jsonl
   

4. Report record counts per split, label distribution, and any rejected records.

Do not commit any file under data/raw/ or data/processed/ — only the manifest under data/manifests/.