我有一个想法,即在当前项目中为非结构化数据实现线索功能,类似于Apache Atlas或OpenLineage中的结构化数据线索。

2.1 Data-Juicer's existing capabilities inventory

Level	Existing assets	Completeness
Sample-level tracking	Tracer / RayTracer	✅ Complete
Event logging	EventLoggingMixin + 8 types of EventType	✅ Complete
Pipeline topology	PipelineDAG	✅ Complete
Content hash	HashKeys.hash/minhash/simhash/imagehash/videohash	Assets exist but not linked
Agent scene metadata	agent_request_model / agent_pt / agent_request_id	Partial scenes
File-level provenance	_data_source_provenance_block (in HTML report)	Report-level
Dataset/field-level lineage	❌ None	❌ Need to be built
Asset-level lineage	❌ None	❌ Need to be built
ZZTERM5ZZ call lineage	❌ None	❌ Need to be built
Compliance chain	❌ None (though there are PII/NSFW algorithms)	❌ Need to be built
Equivalent deduplication lineage	❌ None	❌ Need to be built

2.2 Target capability matrix