This approach is based on empirical data, is robust to error, and has high explanatory power (all sensitivity and error-complexity figures are accessible via the description of the symbolic regression evolutions). However, to capture the full scope of causality, the "inherited fragility", the AI makes use of signal processing and other techniques. Figure 5 depicts the use of the wavelet coherence method, which has been previously applied to the study commodities and financial time series \cite{el_Alaoui_2015} \cite{Nagayev_2016}, to understand when and how strongly an off-chain signal (in this case, the usage of a popular Ethereum web wallet service) affects the price of the cryptocurrency, ether. The AI uses this not only for disambiguation but to actually map the causal relationship in time and frequency domain (i.e. learn when one signal lags or leads the other).