Comment by 1a1a11a

Thank you! Yes, this addresses the scan-resistant issue, but I am not sure how different it would be on workloads without scan. But the workloads we used in evaluations have very few scans, and the power of SIEVE comes from quickly evicting newer objects so I guess "random two" would not improve. But I do not have a number right now, I am happy to add this to the comparison list.