Where ndenote the complexity boundaries; we integrate over time, since we are measuring usage per month. The outflows are implied, and not shown in the graph, but we assume that whatever attention BitcoinCash is losing, Bitcoin is gaining --although in practice there might be as well leakages towards other competing forks.
So, at complexity level 11 (the worst error is what matters) BCH returns are driven by inflows into the BitcoinCash economy (2).
\(BitcoinCash_1=(-0.13+\frac{4.27\cdot10^{-6}}{github})\).
And outflows can be described by the Bitcoin economy gains (3). 
\(Bitcoin_1=(4.80+\frac{2.90\cdot10^{-2}}{duckDuckGo})\).
At the same level of complexity the error measure associated to the Bitcoin model (0.04) is lower than for the BitcoinCash model (0.283); again, the result demonstrates the behavioral traits of the economic agents, as you would expect attention flows towards a software code repository (Github) become a factor for the newer coin, while the more established coin has higher visibility in organic channels (in this case, duckDuckGo, a search engine popular among developers).
This formulation encapsulates tacit knowledge since the model includes information in people's heads (e.g search patterns are revealed preferences, but are private to the user until the data is mined). It also contains explicit knowledge: blockchain unprecedented advantage is the public availability of transactional data. But from an investment perspective, the reason why modeling  the level of trust is important is because the shape of the trust surface has a relationship with the probability of gain or loss  \cite{J_sang_2004} ; this extends as well to the domain of computational trust \cite{J_sang_2016} a discipline in information security that deals with the analysis of trust structures such as those of a PKI (Public-Key Infrastructure) .
Fields finance. Another way to analyze the condition of asymmetry is by looking at trust imbalances among the same set of variables. In this way, we force the evolutionary algorithm to choose the best model that simultaneously contains both variables, and that allows for the flow to be visualized on a higher dimensional space (e.g. a vector field). To make the streams fully descriptive of the path to material economic activity (not simply market sentiment) we use blockchain fees rather than returns, and time series of daily usage data rather than share of inflows; the off-chain data expressively includes variables related to transactional activity (e.g. cryptocurrency exchanges, cryptocurrency payment platform for merchants). This allows for a better description of the causal relationship, and facilitates additional verification using forecasting methods such as bivariate Granger causality \cite{Seth_2007}
The resulting inflow equations are arranged into a field of the form given by (4).
\(\left\{Fees_{BCT}\ ,\ Fees_{BCH}\right\}=\left\{f\left(X_1,\ X_2\right),\ g\left(X_1,\ X_2\right)\right\}\).
Where X1 refers to huobi.pro, a Chinese exchange; X2 refers to coinpayments.net, a payments platform.
We slice the data by month (from September to November), to focus on the periods of analysis that are of interest -- where we want to study the persistence or the break of trust symmetry. We obtain 6 equations in total, 2 for each month (each one describes how usage of the services under study may predict the movements in BTC or BCH fees). To obtain the rate of change of inflows levels (rather than levels themselves) we use the expression (1) applying a derivative at both sides and without other modification than assuming outflows equal 0; this requires that we compute the gradient of the field. The results are plotted in Figure 14, where X1 is the component in the horizontal axis.