Based on this behavior, the two scenarios for Ristretto would be: 1. If the chosen sample keys have high Estimate, the incoming keys could be rejected, affecti

Based on this behavior, the two scenarios for Ristretto would be: 1. If the chosen sample keys have high Estimate, the incoming keys could be rejected, affecting the Hit ratios. However, our Hit ratio benchmarks didn't show such behavior (within 1% of exact LFU). > With this approach, the hit ratios are within 1% of the exact LFU policies for a variety of workloads. 2. The chosen keys have low Estimates, in which case they'd be removed and hence, won't suffer from repeated selection. So, yeah. Doesn't affect Ristretto. But, good thing to keep in mind for other workloads.