Through purposeful introduction of malicious transactions (tracking transactions) into randomly select nodes of a (database) graph, soiled and clean segments are identified. Soiled and clean measures corresponding those segments are then computed. These measures are used to repose the problem of critical database elements detection as an optimization problem over the graph. This method is universally applicable over a large class of graphs (including directed, weighted, disconnected, cyclic) that occur in several contexts of databases. A generalization argument is presented which extends the critical data problem to abstract settings.