Memoization is deterministic caching, meaning that the structure is a perfect stand-in for a deterministic function. The "determinism" of the function itself is relative to some agreed upon lifespan, during which if you feed it some input, the corresponding output is guaranteed to always be the same (e.g. until the next reboot, the runtime of the main program, the span of a function call, etc). Within that timespan, the result of computations can be reliably stored and retrieved from the cache. When it's exceeded, the function's corresponding cache also goes stale and should be considered unreliable and to be discarded.

Note that in the case of memoization, the guarantee of determinism (and associated lifespan) is attached to the function, not the caching structure. Unlike some other caching approaches that rely on a replacement policy.