size_lru achieves 1.85M/s effective throughput (100% baseline) - highest among tested libraries. The LHD algorithm asks 'Which item has the lowest expected hits per byte?' not 'least recently used?' LHD quantifies hit density = expected_hits/size via 16 age classes and 4096 buckets. Random sampling of 64 candidates achieves 69.83% hit rate vs 54-55% for LRU with O(1) operations.

Embedding version numbers and demo code in readme.md is a very common requirement.

There are some similar tools on the market, but none of them works well. So I wrote one myself ( executable size 200 KB ).