The algorithm is working on a tree, so we need to have
all FinDocs in a hash first (otherwise the algorithm
would become slow). Currently we fetch all FinDocs
of the FinDoc's project. This would have to be
extended to really fix this bug.