I think you're misunderstanding how the assignment works. Centrifuge attempts to assign each read to the _most specific_ taxon possible. Only if the read cannot be assigned to a taxon unambiguously does it assign the read further up the tree. Your output is showing that you have many reads assigned to Homo sapiens at the species level, and 12 reads assigned to Homo at the genus level. That means these reads may have been compatible with H. sapiens, but matched equally well with one or more other species within the Homo genus. The counts up the taxonomy are not cumulative --- if you wanted to know all reads coming from Homo at the genus level or lower, you would sum the counts at the genus itself (node 9605) with all of the counts on nodes that are descendants of Homo.