Analysis of metabolic networks through characterising flux vectors such as Elementary Flux Modes and Extreme Pathways has shown to be valuable in many application within systems biology. However, their application is still limited to medium-scale metabolic networks due to the combinatorial explosion of the set size. Enumeration of the full set in genome-scale metabolic networks remains impossible, but few techniques exist for partial enumeration. Currently, an adapted version of the Canonical Basis Approach for generation of Extreme Pathways has allowed partial enumeration with good sampling quality through a filter. However, computational efficiency currently limits its application, especially when moving towards larger networks. One approach to increase this efficiency significantly is through candidate narrowing based on bit pattern trees. Here, many of the candidate modes are removed every iteration based on their ability to pass a rank test. However, the depth of the trees should be adapted based on the filter setting and the current amount of candidates to process. In this work, a novel algorithm is presented using candidate narrowing in a stochastic partial enumeration of Extreme Pathways. A case study for the central carbon metabolism of Escherichia coli showcases the adaptive tree depth and the overall performance increase for different settings.