add train_V2_full.jsonl — single canonical file: cleaned base + targeted category-complete rows, 741,803 rows, 0 invalid, all categories, targeted supersedes on label conflicts

v3: verified sanitized base (691,675 rows), category-complete targeted set (50,203 rows), audit tooling, README

Upload data/train_V2_original.jsonl with huggingface_hub

Upload data/train_V2_fixed.jsonl with huggingface_hub