but i still suspect that we haven’t actually nailed down via profiling the bottleneck in that method