We examine (directed) greybox fuzzing from a geometrical perspective, viewing dissimilarities on inputs and on control flow graphs (with dynamical statistics) as primitive objects of interest. We prototype and evaluate GoExploreFuzz, a greybox fuzzer for time-intensive programs that incorporates this perspective. The results indicate useful capabilities for greybox fuzzing that have hitherto been underutilized, notably quantifying the diversity of paths and autonomously tuning the "bandwidth" of mutations.