Research and education on Sindhi NLP, and as a starting point for sentiment tools on short, single-sentence Sindhi text. - Input is expected to be Sindhi in Arabic script, cleaned with the project's cleantext (Unicode NFKC, invisible characters removed, whitespace normalized) before prediction. - Decisions about people (employment, credit, moderation or legal outcomes) without human review. - Text far from the training data: long documents, social-media slang or code-mixed text, other languages or other Sindhi scripts, sarcasm-heavy text. These are untested. - Measuring sentiment of a population or topic without checking accuracy on that kind of text first. Natural dataset (human-labeled).…