I assume this doesn't improve perf much because you pretty quickly zoom in on a slice of hashes that are not evenly distributed. 37GB feels like a lot of hashes

I assume this doesn't improve perf much because you pretty quickly zoom in on a slice of hashes that are not evenly distributed. 37GB feels like a lot of hashes but it's pretty small compared to the space of all hashes, I suspect this approach would be more effective if we were searching over a much bigger chunk of data.