To effectively answer the large number of authorization requests per second, Google's Zanzibar project introduced a novel way of caching that reuses authorization results. In SpiceDB, this caching mechanism is used to enable super large scales by breaking down check requests into subproblems, which can be cached and reused while answering other check queries. However, this approach introduces the issue of staleness, where cache entries become invalid if underlying data changes. To address this, Zanzibar settled on a configurable temporal window for caching, throwing out all entries after that window closes. The system also uses token hashring to distribute the cache across nodes in a stable fashion, ensuring maximal reuse and deduplication of requests. As a result, SpiceDB users can often see high cache rates, with up to 60% over all subproblems being computed, and up to a 40% reduction in the number of requests. The caching strategy is designed to be configurable and adaptable to different use cases, with plans for future updates and improvements.