Extending Granger causality to separate common cause in some cases

Extending Granger causality to tell a direct link from a hidden common cause: the earlier signal is whitened on the joint past, then a rank-based HCR test of residual independence with calibrated p-values is applied, which works when the innovations are non-Gaussian. Includes a one-file Python tool, a browser version, and checks on synthetic data, rivers, stock returns and microservice incidents.