Researchers propose a new framework for evaluating how time-series classifiers connect their own classes to each other. They introduce counterfactual transition graphs (CGTs), which use edge weights to represent the reliability of transitions between class prototypes. On a six-class hand-movement task, the CGT reveals a non-trivial topology that contradicts the binary confusion matrix. The study finds that gradient-based methods can reach almost any class by stepping off the data manifold, while replacement-based methods stay on it and fail on rigid boundaries. --- Why it matters: This matters to researchers in AI because it provides a new way to evaluate the structural relationships between classes in time-series classifiers, which is crucial for diagnostic interpretability. The study's findings also highlight the limitations of existing counterfactual explanation methods. Source: https://arxiv.org/abs/2608.23164