Refresh extract: preserve Vietnamese noun 'Lỗi' in titles + widen decorative-quote stripping (boundary/run policy, word-internal place-name apostrophes preserved); migrate hf_export to the per-doc JSONL tier

Declare embed + reduce as non-default configs (keep main table = documents/parse)

Document embed + reduce stages in dataset card

Add embed + reduce stages: 147K-doc dense vectors (2048-d) + PCA/t-SNE/UMAP + HDBSCAN cluster ids

May-2026 recovery: live-API body refresh + NULL-markdown semantics + title/agency sweep

Cap Chưa phân loại visual size so the informative tail is readable

Document markdown junk-strip rule in dataset card

Vietnamese National Legal Database (vbpl.vn) corpus with structure + embedding layers

Rework citation: English title + Vietnamese site title as author

Vietnamese National Legal Database (vbpl.vn) corpus with structure + embedding layers

Vietnamese National Legal Database (vbpl.vn) corpus with structure + embedding layers

Vietnamese National Legal Database (vbpl.vn) corpus with structure + embedding layers

Vietnamese National Legal Database (vbpl.vn) corpus with structure + embedding layers

Vietnamese National Legal Database (vbpl.vn) corpus with structure + embedding layers

Vietnamese National Legal Database (vbpl.vn) corpus with structure + embedding layers

Shard documents.parquet into 32 files to fix HF dataset viewer JobManagerCrashedError

Drop monolithic documents.parquet (switching to sharded layout to fix viewer JobManagerCrashedError)

Drop t-SNE plots from dataset card (UMAP-only viz)

Drop orphaned t-SNE PNGs (dataset card now UMAP-only)

Vietnamese National Legal Database (vbpl.vn) corpus with structure + embedding layers

Vietnamese National Legal Database (vbpl.vn) corpus with structure + embedding layers

initial commit