The Crimson Hexagonal Archive Hugging Face Dataset: Work Plan v3 is a metadata witness for a classifier-centered dataset methodology.
The plan supersedes a basic export model and a decision-tree approach by placing an automated classifier at the center of corpus preparation. The classifier is assigned three simultaneous tasks: identifying provenance mode, identifying artifact mode, and proposing heteronym attribution.
The methodology preserves original attribution and classifier-proposed attribution in parallel. A prediction does not silently rewrite the source record. Confidence tiers route uncertain cases to manual review.
The plan distinguishes two tracks. The first creates a dataset-internal research layer. The second concerns deliberate propagation of reviewed classifications into public metadata systems such as repository records or Wikidata.
It also proposes ablation sets with text-only, minimal-header, and provenance-header variants; separate artifact-level and chunk-level configurations; and a model-collapse experiment comparing provenance-bearing and provenance-stripped corpora.
The body at #1001 is not the full work plan. It is a DataCite capture paired to the complete in-archive version at deposit #745.