正文
获取原文件references/dataset-curation-policy.md
D015 curates container structure; it does not interpret medicine.
Allowed bounded processing:
- inventory;
- one-level ZIP/TAR/GZIP extraction/decompression;
- hash/size verification;
- explicit member-role assignment;
- copying curated members to the handoff package.
Blocked:
- recursive/nested archive extraction;
- symlinks/hardlinks/devices/FIFOs;
- path traversal;
- encrypted ZIP members;
- unbounded decompression;
- auto-executing notebooks/scripts/SQL;
- guessing clinical table roles from names alone.
Roles:
- clinical_data
- reference_table
- data_dictionary
- documentation
- ignore
- review
D001 re-inventories the curated handoff. D003 still decides privacy/model visibility.