We are concerned with measuring the maximum distance (infinity-norm) between any one of the variables. We see that most data points have larger metrics, with a few points breaking the symmetry (e.g. 0.8-1 bin). Since we are interested in material connections between the asymmetry of the neighborhood data space and actual trust asymmetry, we use fees as response variable (an actual on-chain transaction metric), alongside the behavioral signals of off-chain economic and investment activity.  
 

Mathematical invariants

Ensembles of models of diverse but comparable performance and complexity lead to a trust metric \cite{Vladislavleva_2015}. Figure 7 shows how the intelligent agent perceives the trustability of the prediction, what may happen in regions of unknown parameter space (when it is exposed to unseen data) or when the underlying system changes. The AI naturally finds interesting those inputs that show invariance, as well as the points where the symmetry breaks for the others.