A benchmark dataset for evaluating prompt strategy sensitivity in Kannada, a low-resource Dravidian language.
Language: Kannada (kn)
Tasks: Sentiment Analysis (100), Question Answering (75), Summarization (50)
Total: 225 culturally grounded samples
Inter-annotator agreement: Cohen's κ > 0.80
Each sample contains: id, task, input_text, label, difficulty, domain.
