When backtracking there may be more than one starting MCVX that leads back to the same required point. This could lead to double counting if you scan the data structure and start the backtrack search from every MCVX. One way to reduce the risk is to place a cut on scanning and only select sink vertices; that way you know you are starting with a different track each time. Of course, even that won't prevent double counting if you backtrack far enough. If you backtrack all the way back to the source vertex then every track that it produced directly or indirectly, will lead back to it! So think carefully about what you want to do.