Interpretable A|posteriori error indication for graph neural network ...