yeah, incremental tracking is sort of what obviates the need for cold BUILD file parsing optimization.. but what if the BUILD file parsing could happen concurrently in e.g. <8 seconds vs ~80 in the cold case.