Problem

scripts/curate_pai_samples.py and scripts/download_pai.py both accept a start-end range form for chunk IDs and both implement it as range(start, end) (exclusive of end).

download_pai.py spells this out in its --chunk-ids help string:

range '0-3' (exclusive end, downloads 0,1,2)

curate_pai_samples.py does not, and its docstring example actively misleads:

"""
example:
python scripts/curate_pai_samples.py \\
  --clip-index-path /path/to/PAI_datset/clip_index.parquet \\
  --chunk 3116-3119 --num-samples 16 \\
  ...
"""

A user following that example reasonably expects 4 chunks (3116, 3117, 3118, 3119) but silently gets 3 (3116, 3117, 3118), because:

def _parse_chunk_ids(chunk_arg: str) -> list[str]:
    if "-" in chunk_arg:
        start_s, end_s = chunk_arg.split("-", 1)
        start, end = int(start_s.strip()), int(end_s.strip())
        return [str(i) for i in range(start, end)]

Fix

Pure docs / UX change. The _parse_chunk_ids implementation is unchanged.

• Update the --chunk argparse help to mirror download_pai.py's wording and explicitly call out the exclusive-end behavior, with a cross-reference so the two stay in lockstep.
• Adjust the docstring example from --chunk 3116-3119 to --chunk 3116-3120 and add an inline note clarifying that 3120 is excluded. The new example still selects 4 chunks (3116-3119), matching the obvious reading of the original example.

Verification

No code logic touched. New help text correctly describes existing behavior of _parse_chunk_ids.