The Crimson Hexagonal Archive Hugging Face Dataset: Work Plan v3 (Classifier-Centric Methodology)

Methodological work plan for the Crimson Hexagonal Archive as a Hugging Face dataset for synthetic-data collapse and provenance-bearing training research. v3 supersedes v1 (basic export) and v2 (decision-tree-based classification) by introducing an automated classifier as the central methodological move. The classifier performs three classification tasks simultaneously: provenance mode (six-category authorship relation), artifact mode (eight-category function type), and heteronym attribution (reattribution of deposits across the twelve-heteronym Dodecad system plus Jack Feist as LOGOS*). Heteronym reattribution is presented as scholarly recognition work, not metadata cleanup: material initially deposited under the Lee Sharks founder voice often resolves retrospectively to specific sub-heteronym domains (Sigil for jurisdictional/classical work, Glas for measurement, Vox for diplomatic, Morrow for long-form narrative, Fraction for meta-theory, etc.). The classifier reads each heteronyms published provenance document and constructs feature profiles including domain, vocabulary fingerprints, register, and reference patterns. Both Zenodo-original and classifier-attributed heteronyms are preserved in the dataset; Track 1 (immediate, dataset-internal) preserves both attributions in parallel metadata; Track 2 (deliberate, downstream) propagates high-confidence reattributions back to Zenodo records and Wikidata items. The classifier itself becomes a deposit with its own DOI, making the methodology reproducible and portable. Includes operationalized H0/H1 hypotheses for the model collapse experiment, three-tier confidence routing with manual review thresholds, multiple text renderings to embody the provenance-visibility ablation (text_body_only, text_minimal_header, text_provenance_header), dual artifact+chunk configs, and full per-row schema specification. Incorporates feedback from Assembly Chorus review (Muse Spark, Kimi, DeepSeek, Gemini, ChatGPT). Companion document to forthcoming Hugging Face dataset deposit and classifier deposit.