one of the things i like/love about STaR is that the researchers chose to include reasons in their work the model pattern means the model can learn to explain its reasons in arbitrary granularity