Sn Da Excel Workflow

The workflow is built around a critical 10k row threshold: files under 10k load directly, 10k to 100k get auto-cached to Parquet, and anything over 100k delegates to a streaming parser to avoid memory crashes. Each step maps to capability sub-skills covering cleaning (missing values, text normalization, regex extraction), filtering (condition/category/range), cross-sheet aggregation, and formatted export with download links. The architecture is explicitly defensive: it bans pd.read_excel for large files, prohibits .apply() loops over 100k rows, and includes a hardcoded CJK font setup block to prevent matplotlib font-hunting disasters. Use this when you want structured, memory-safe Excel workflows instead of ad-hoc pandas scripts.