What it is
A cache that is full must decide which entry to evict. The best policy depends on how requests are distributed: LRU adapts to recency, LFU to popularity, and newer designs such as ARC, W-TinyLFU and S3-FIFO try to get both while resisting scans. This simulator replays one request trace against several policies at once, draws the hit ratio against cache size (the LRU curve in a single pass), and lets you replay any policy request by request.
When to use it
Use it to see why a one-time scan flushes an LRU cache, why a loop slightly larger than the cache gives LRU and FIFO a zero hit ratio, why LFU is slow to follow a moving hot set, and how far each policy is from the offline optimum. Paste your own key sequence to test a trace you care about.
Gotchas & limits
Objects have unit size and there is a single cache: no byte sizes, TTLs, concurrency, write costs or prefetching. The policies are implemented from their published descriptions and checked with properties (for example, nothing beats the offline optimum), but they have not been compared against reference implementations, and real caches add many engineering details. Results show the mechanism, not a prediction for your system.