The pragmatic frames of spurious correlations in machine learning: Interpreting how and why they matter | Synapse