KANDINSKYPatterns our Swiss Knife for studying explainbale AI are mathematically describable, simple self-contained hence controllable test data sets for the development, validation and training of explainability in artificial intelligence.